Health Consultation HUDSON RIVER PCBs

RESIDENTIAL PROXIMITY TO THE HUDSON RIVER AND HOSPITALIZATION RATES FOR

ISCHEMIC HEART DISEASE AND STROKE 1990-2005

Westchester Rockland Putnam Orange Dutchess Ulster Columbia Greene Rensselaer Albany Washington and Saratoga Counties

New York

EPA FACILITY ID NYD980763841

Prepared by The New York State Department of Health

JUNE 16 2011

Prepared under a Cooperative Agreement with the US DEPARTMENT OF HEALTH AND HUMAN SERVICES

Agency for Toxic Substances and Disease Registry Division of Health Assessment and Consultation

Atlanta Georgia 30333

Health Consultation A Note of Explanation

A health consultation is a verbal or written response from ATSDR or ATSDRrsquos Cooperative Agreement Partners to a specific request for information about health risks related to a specific site a chemical release or the presence of hazardous material In order to prevent or mitigate exposures a consultation may lead to specific actions such as restricting use of or replacing water supplies intensifying environmental sampling restricting site access or removing the contaminated material

In addition consultations may recommend additional public health actions such as conducting health surveillance activities to evaluate exposure or trends in adverse health outcomes conducting biological indicators of exposure studies to assess exposure and providing health education for health care providers and community members This concludes the health consultation process for this site unless additional information is obtained by ATSDR or ATSDRrsquos Cooperative Agreement Partner which in the Agencyrsquos opinion indicates a need to revise or append the conclusions previously issued

You May Contact ATSDR Toll Free at 1-800-CDC-INFO

or Visit our Home Page at httpwwwatsdrcdcgov

Health Consultation A Note of Explanation

A health consultation is a verbal or written response from ATSDR or ATSDRrsquos Cooperative Agreement Partners to a specific request for information about health risks related to a specific site a chemical release or the presence of hazardous material In order to prevent or mitigate exposures a consultation may lead to specific actions such as restricting use of or replacing water supplies intensifying environmental sampling restricting site access or removing the contaminated material

In addition consultations may recommend additional public health actions such as conducting health surveillance activities to evaluate exposure or trends in adverse health outcomes conducting biological indicators of exposure studies to assess exposure and providing health education for health care providers and community members This concludes the health consultation process for this site unless additional information is obtained by ATSDR or ATSDRrsquos Cooperative Agreement Partner which in the Agencyrsquos opinion indicates a need to revise or append the conclusions previously issued

You May Contact ATSDR Toll Free at 1-800-CDC-INFO

or Visit our Home Page at httpwwwatsdrcdcgov

HEALTH CONSULTATION

HUDSON RIVER PCBs

RESIDENTIAL PROXIMITY TO THE HUDSON RIVER AND HOSPITALIZATION RATES FOR

ISCHEMIC HEART DISEASE AND STROKE 1990-2005

Westchester Rockland Putnam Orange Dutchess Ulster Columbia Greene Rensselaer Albany Washington and Saratoga Counties

New York

EPA FACILITY ID NYD980763841

Prepared By

The New York State Department of Health Center for Environmental Health

Troy New York Under cooperative agreement with

The US Department of Health and Human Services Agency for Toxic Substances and Disease Registry

Division of Health Assessment and Consultation Atlanta Georgia

For additional information about this document you may contact the

New York State Department of Health Center for Environmental Health

547 River Street Flanigan Square Room 300 Troy New York 12180-2216

1-518-402-7950 or

1-800-458-1158

E-mail beoehealthstatenyus

For additional information about this document you may contact the

New York State Department of Health Center for Environmental Health

547 River Street Flanigan Square Room 300 Troy New York 12180-2216

1-518-402-7950 or

1-800-458-1158

E-mail beoehealthstatenyus

Table of Contents

Executive Summary helliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphellipii

1 Introduction helliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphellip 1

2 Background helliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphellip 3 Cardiovascular Disease Hudson River PCB Contamination

` POPs and PCBs PCBs and Cardiovascular Disease Health Outcomes

3 Follow-up Study Data Sources and Methodshelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphellip8 Study Area and Exposure Areas

Hospitalization Data Statistical Analysis

4 Results helliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphellip 13 Demographics Associations with Proximity to the River

Statistical Analysis Proximity to the River and CVD Hospitalization Risk Summary of Results

5 Discussion helliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphellip 19

6 Conclusion and Recommendations helliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphellip25

7 Agency Information (Preparers of the Report) helliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphellip 27

8 References helliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphellip 28

9 Tables helliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphellip 33

10 Figures helliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphellip 44

11 Appendices helliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphelliphellip 48

EXECUTIVE SUMMARY

Background In October 2006 Member of Congress Maurice Hinchey requested that ATSDR conduct a health consultation to assess cardiovascular disease among people residing near the Hudson River A 200-mile segment of the Hudson River from Hudson Falls south to New York City is a National Priorities List (NPL) (Superfund) site due to contamination of the river with polychlorinated biphenyls (PCBs) from GE sites in Hudson Falls and Ft Edward In support of his request Member of Congress Hinchey cited three studies of cardiovascular hospitalizations conducted with similar data and methods These 2004-2005 statewide studies suggest residents of ZIP codes containing hazardous waste sites with persistent organic pollutants (POPs) such as PCBs have elevated risk for hospitalizations for cardiovascular disease including ischemic heart disease stroke and hypertension Analyses conducted separately for ZIP codes next to the Hudson River suggest residents of these specific ZIP codes also have increased rates of hospitalization for cardiovascular disease

In response to Congressman Hincheyrsquos request for follow-up of the 2004-2005 findings for the Hudson River area ATSDR and NYS DOH agreed to conduct a similar group-level study but at a different level of geography the Census block group By using a level of geography that is generally smaller in area and population than a ZIP code this follow-up seeks to more accurately adjust for socio-economic and other factors that affect cardiovascular hospitalizations and to improve upon the geographic classification of relative distance to the Hudson River The follow-up investigationrsquos analyses are restricted to the twelve counties abutting the Hudson River from Washington County south to Westchester County Within the study arearsquos twelve counties cardiovascular hospitalization rates in areas closest to the Hudson River are compared with rates in areas farther from the river

Objectives The follow-up investigationrsquos specific objectives are to increase the studyrsquos ability to control for confounding by socio-economic status through the

use of block groups rather than ZIP codes as the unit of analysis compare hospitalization rates among residents of block groups nearest the Hudson River to

those among residents of block groups farther from the Hudson River and compare the findings of this review to those of the earlier studies

Methods This study examines CVD hospitalization rates and residence near the Hudson River in the 12 counties abutting the Hudson River below Hudson Falls from 1990 through 2005 The hospitalization data for ischemic heart disease and stroke were acquired from the NYS DOH Statewide Planning and Research Cooperative System (SPARCS) established in 1979 to collect detailed records on discharges from hospitals located in New York State Statewide US Census data at the block group level from 1990 and 2000 were used to calculate the number of people in each race sex and age group for the study years

Classification of the population into potential exposure categories was based exclusively on location relative to the Hudson River Two types of exposure proxy categories were developed block group adjacency and block group centroid distance to the river shore For the adjacency categories the hospitalizations were assigned to the adjacent category if the residential address was in a block group directly abutting the Hudson River For the distance categories block

ii

group centroid (population-weighted) distance to the river shore was used to assign hospitalizations in each block group to one of three categories less than one-half mile from the Hudson River (closest) between one-half to one mile (close) from the river and farther than one mile from the river (distant) Multivariable regression analyses were conducted to compare rates of hospitalizations among the exposure categories assigned at the block group level These analyses controlled for age sex and race using individual-level data and population density percent Hispanic median income education and distance to the nearest hospital at the block group level

Results The use of block groups resulted in 2371 block groups as the units of analysis for the study rather than the 368 ZIP codes in the 12-County study area The average population per block group in the study area is 1200 compared to average population of 8000 per ZIP code The smaller geographical area and smaller populations in block groups contributed to improving the accuracy of the assignment of income and other group-level demographics to the hospitalizations included in the study and enhancing control for confounding by socio-economic status (Tables 8a and 8b)

The studyrsquos multivariable analyses showed relatively small but statistically significant elevations of CVD hospitalization risk for residents of block groups categorized as adjacent in close and closest proximity to the Hudson River There is no evidence that residents of the closest block groups (less than frac12 mile) had higher hospitalization rates than residents of the close block groups (between frac12 to 1 mile) If there were a strong association between exposures from living near the river and the risk of CVD hospitalization such a ldquodose-responserdquo might have been shown by the analyses

The data showed strong statistically significant associations between CVD risk and socioshyeconomic status To enhance the control for socio-economic status separate analyses for each income quartile were conducted These stratified analyses showed that residence in block groups near (close closest or adjacent to) the Hudson River was associated with increased risk for CVD hospitalizations among residents of lower income block groups Conversely residence in block groups near the Hudson River was associated with decreased risk for CVD hospitalizations among residents of higher income block groups In other words in analyses that held income constant by looking only within the lower or higher income populations living near the river shows opposite effects depending upon which income group is considered The estimated increased and decreased risks are of sufficient magnitude to cancel each other out when average rate ratios for the four income quartiles are calculated The average rate ratios can be interpreted as showing no evidence that hospitalizations among the general population are elevated for residents living near the river

This type of group-level statistical analysis is not able to precisely identify the reason for these differing results However the strength and direction of associations among variables in the statistical models can be evaluated in light of known risk factors to suggest likely explanations In this case socio-economic status appears to be the most likely explanation for the findings The strong association between socio-economic status and residential location (the indicator of potential exposure) and between socio-economic status and CVD hospitalization risk makes it likely that the small but statistically significant elevations shown in the multivariate models

iii

(Tables 8a amp 8b) are the result of the inability to completely control for socio-economic statusrsquo effects on CVD hospitalization risk The strong effect of socio-economic status on residential location (exposure proxy) and on CVD hospitalization risk makes it likely that these overall multivariable regression results are affected by uncontrolled ie residual confounding by socioshyeconomic status

Because of the similarities in study design and statistical analysis the prior studyrsquos findings can be compared to some degree with the findings of the multivariate regression analyses in the current study The major difference between the studies is the composition of the comparison areas for the analyses Because of this difference as well as other differences described in the report the studies are not entirely comparable

CVD hospitalization rate ratios estimated for living in block groups adjacent closest or close to the river in the current study are substantially lower than the rate ratios for residence in ZIP codes adjacent to the Hudson River in the previously published studies In most cases the current analyses show elevations below 10 in contrast to elevations of 20 to 39 in the prior studies The results of the current study are quite similar to the prior study results in terms of the associations reported for age sex race and income These similarities and the general methodological similarities between the current and prior studies suggest the likelihood that the prior studiesrsquo findings also were affected by confounding by socio-economic status

Conclusions and Recommendations While this type of study can neither prove nor disprove causal links among risk factors and health effects it is useful for examining evidence for unusual patterns that might warrant additional investigation This evaluation of hospitalization data residential location and socio-economic factors in block groups in the 12 Counties abutting the Hudson River revealed a closely coinciding pattern of lower socio-economic status and residential proximity to the Hudson River the proxy for potential exposure and a strong association of lower socio-economic status and risk for CVD hospitalizations The strong associations among residential location socio-economic status and CVD risk prevent us from drawing definitive conclusions from the multivariable regression analyses that showed modest but statistically significant elevations of CVD hospitalization risk for residents of block groups near the river

To draw more definitive conclusions particularly for outcomes such as CVD that are determined by multiple known risk factors studies seeking to investigate the role of PCB or POP exposures will require resource-intensive methods that include gathering individual-level exposure or biomonitoring information as well as medical histories and information on other risk factors

This current investigation did not include environmental or biological sampling data One study did gather such data in the Hudson River region with the highest potential exposures (Hudson Falls Fort Edward) This study detected a statistically significant reduction in the total PCB concentrations in outdoor air samples at a distance of 1200 meters (34 mile) from the river However the PCB levels in outdoor air in this area were lower than those in other communities with known PCB-contaminated sites and similar to levels reported in other locations in the northeastern United States The study found no detectable differences in study subjectsrsquo serum

iv

PCB levels associated with proximity to contaminated portions of the river or other PCB sources (Fitzgerald 2007 Palmer 2008)

It is recommended that the results of the current investigation of CVD hospitalization rates and residence near the Hudson River be shared with NYS residents and other stakeholders with interest in this issue and be submitted for publication in a peer-reviewed journal

v

1 INTRODUCTION

The New York State Department of Health (NYS DOH) has a cooperative agreement with the Agency for Toxic Substances and Disease Registry (ATSDR) to perform environmental exposure and health assessments conduct health statistics reviews and perform epidemiological studies of populations in New York State which may have been exposed to environmental contaminants In October 2006 Member of Congress Maurice Hinchey requested that ATSDR conduct a health consultation to assess cardiovascular disease among people residing near the Hudson River A 200-mile segment of the Hudson River from Hudson Falls south to New York City is a National Priority List (NPL) (Superfund) site due to contamination of the river with polychlorinated biphenyls (PCBs) from GE sites in Hudson Falls and Ft Edward (See Appendix A for Member of Congress Hincheyrsquos letter)

In support of his request Member of Congress Hinchey cited three studies (Sergeev and Carpenter 2005 [ischemic heart disease] Shcherbatykh et al 2005 [stroke] Huang et al 2006 [hypertension]) These statewide studies suggest residents of ZIP codes containing hazardous waste sites with persistent organic pollutants (POPs) such as PCBs have elevated risk for hospitalizations for cardiovascular disease including ischemic heart disease stroke and hypertension Analyses conducted separately for ZIP codes next to the Hudson River suggest residents of these specific ZIP codes also have increased rates of hospitalization for cardiovascular disease

These studies published in 2005-2006 are ecological studies In an ecological study data are analyzed by evaluating risk factors and health outcomes among groups For example rates of disease are compared for populations of ZIP codes with varying median income levels to see if income levels appear to be related to disease outcome rates In these previously conducted studies rates of hospitalization for residents of ZIP codes with hazardous waste sites containing POPs were compared with hospitalization rates for residents of ZIP codes without such waste sites to look for evidence that potential exposures might be impacting health outcomes The studies report statistically significant elevations of cardiovascular disease hospitalizations in ZIP codes with POPs For coronary heart disease and a subset acute myocardial infarction the study showed elevations of approximately 15 and 20 respectively (Sergeev and Carpenter 2005) for stroke a 15 elevation (Shcherbatykh et al 2005) and for hypertension a 19 elevation (Huang et al 2006) in ZIP codes containing waste sites with POPs compared with hospitalization rates in other New York State ZIP codes (excluding NYC) that did not contain waste sites with POPs (Table 1)

In sub-analyses in these studies cardiovascular hospitalization rates for ZIP codes adjacent to the Hudson River were compared with cardiovascular hospitalization rates for ZIP codes in the rest of the state (excluding NYC) identified as not having inactive hazardous waste sites with POPs These sub-analyses showed statistically significantly elevated hospitalization rates of 36 and 39 for coronary heart disease and the subset acute myocardial infarction a 20 elevation for stroke and a 14 elevation for hypertension (Table 1) The Sergeev and Carpenter Shcherbatykh et al and Huang et al articles state that the Hudson River ZIP codes are in an area where people have higher income smoke less exercise more and have healthier diets This information is cited in these articles as evidence that known risk factors related to lower socioshy

1

economic status are likely not the cause for the elevated hospitalization rates in these Hudson River area ZIP codes

In these previously published reports Behavioral Risk Factor Surveillance System (BRFSS) data for entire counties are provided to support statements about the population of Hudson River ZIP codes having higher income and healthier lifestyles However because the ZIP codes adjacent to the Hudson River include only a proportion of the population of the counties abutting the river these county-level BRFSS data may be misleading regarding characteristics of the ZIP codes adjacent to the river (Figure 1) In fact data from the US Census show that the ZIP codes adjacent to the river have lower median income than the counties within which they are located The median income for the total population of the twelve counties adjacent to the Hudson River was $61010 in 2000 while the median income for the ZIP codes adjacent to the river was $45356 This median income for the population of the ZIP codes adjacent to the river is also slightly lower than the statewide (excluding NYC) median income of $47735 (US Census 2000) Thus the Hudson River ZIP codes analyzed in the Sergeev and Carpenter Shcherbatykh et al and Huang et al studies include a population of lower not higher socio-economic status compared to the rest of NYS (excluding NYC)

Because CVD and many CVD risk factors are associated with lower socio-economic status it is especially important to adjust as much as possible for socio-economic status when conducting comparative analyses of CVD hospitalizations or disease incidence or prevalence Regarding the hypothesis about POPs and CVD research studies described in more detail below have suggested possible biological mechanisms for POP exposuresrsquo effects on the heart There is very little research on human exposures to POPs and potential cardiovascular effects however

In response to Congressman Hincheyrsquos request for follow-up of the 2004-2005 findings ATSDR and NYS DOH agreed to conduct a similar study also ecological but at a different level of geography the Census block group By using a level of geography that is generally smaller in area and population than a ZIP code this follow-up seeks to more accurately adjust for socioshyeconomic and other factors that affect cardiovascular hospitalizations and to improve upon the geographic classification of relative distance to the Hudson River In addition rather than comparing hospitalizations among residents of Hudson River area ZIP codes to hospitalizations among residents of all other NYS (exclusive of NYC) ZIP codes with no waste sites with POPs this follow-up investigationrsquos analyses are restricted to the twelve counties abutting the Hudson river from Washington County south to Westchester County Within the study arearsquos twelve counties cardiovascular hospitalization rates in areas closest to the Hudson River are compared with rates in areas farther from the river This more localized approach may also assist with reducing the potential impact of regional differences in medical care practice or reporting factors that could affect the hospitalization data (Ko et al 2007) (Table 2 shows a more detailed listing of the methods used in the prior studies and follow-up study)

Regarding environmental exposures it is important to note that living near a hazardous waste site that contains PCBs or other POPs does not necessarily mean that human exposure from the site has occurred Depending upon the conditions at a site or the specific chemicals exposures may be highly improbable Information on body burden for example PCB levels in blood or on levels of the chemical in off-site air dust or soil is needed to support a claim of exposure The

2

studies published in 2005-2006 did not provide specific information such as environmental or biomonitoring data in support of the assumption that residence in ZIP codes containing hazardous waste sites with POPs or residence in ZIP codes adjacent to the Hudson River resulted in increased exposures to POPs in the population For the purpose of this follow-up we did not conduct environmental or human biological sampling to address this issue Rather this follow-up replicates the prior studiesrsquo approach with some enhancements to better control for confounding particularly confounding by socio-economic status

This descriptive epidemiologic investigation uses already existing hospitalization data from the Statewide Planning and Research Cooperative System and population data from the United States Census to evaluate comparative rates of cardiovascular hospitalizations While this type of investigation can not show proof of cause and effect it can assist in interpreting relationships among indicators of CVD burden and hypothesized exposures associated with distance from the Hudson River This descriptive investigation may also assist in determining whether additional health investigations of these associations are warranted

2 BACKGROUND

Cardiovascular Disease

The widely known risk factors for cardiovascular disease (CVD) in addition to age sex race and family history include tobacco use high blood pressure high serum cholesterol alcohol use obesity low fruit and vegetable intake physical inactivity diabetes low socio-economic status mental ill-health psychosocial stress and the use of certain medications (WHO 2010 NYS DOH 2009 Wing 1988 NYS DOH 2009a)

CVD Definitions CVD is a general term for any disease of the circulatory system The two main types of CVD are ischemic heart disease also termed coronary heart disease and cerebrovascular disease also called stroke Ischemic heart disease (IHD) results from restricted blood flow through the arteries supplying the heart muscle The most serious risk from IHD is acute myocardial infarction (AMI) another term for heart attack AMI occurs when blood flow to the heart is suddenly reduced or stopped usually due to a blockage in a coronary artery Angina is severe constricting pain in the chest often radiating from the area immediately over the heart to the left shoulder and down the arm Cerebrovascular disease refers primarily to stroke the sudden impairment of brain function resulting from interruption of circulation to the brain Ischemic stroke results from blockage of an artery supplying blood to the brain Hemorrhagic stroke occurs due to the escape of blood from a ruptured artery supplying blood to the brain

CVD Burden and Trends CVD is the leading cause of death for men and women of all ages in the United States and in New York State In 1999 for every death from IHD there were five hospitalizations for ischemic disease in NYS (NYS DOH 2009a) Mortality rates from CVD have been declining steadily since the 1980rsquos Research studies estimate that approximately half of the decline is due to improved treatments for CVD and its risk factors such as high cholesterol and high blood pressure and approximately half is due to decreasing incidence of CVD The decrease in incidence is due to trends in health behaviors primarily reduced smoking rates

3

CVD mortality rates do not appear to be declining for all segments of the population however and may actually be increasing among some groups particularly people of lower socio-economic status (Barnett et al 1999) This may be due in part to higher rates of smoking among people of lower socio-economic status as well as to other factors related to socio-economic status and race A growing literature points to multiple causes for CVD and other health disparities including health practices psychosocial stress limited resources discrimination and access to health care (Health United States 2007) Increases in the prevalence of diabetes and obesity in the general population partially offset advances in treatment and favorable trends for other risk factors An increasing ischemic heart disease mortality rate in women ages 35 to 54 since 1997 is attributed to these unfavorable risk factor trends (Ford 2007 Lloyd-Jones 2009) New York State (NYS) and other northeast states have high IHD and low stroke mortality with NYS having the highest IHD mortality and lowest stroke mortality among the 50 states (Howard et al 2009)

CVD Data Sources CVD mortality rates are more easily measured than incidence prevalence or hospitalizations because mortality data from vital records are a comprehensive and relatively accessible data source Information for estimating the incidence and prevalence of CVD outcomes comes primarily from health interview physical examination and record surveys Hospitalization data are available for NYS and the US but are generally collected and analyzed in order to track trends in hospital care and resources expended Actual CVD incidence data validated by physician diagnosis physician examination or reviews of medical records are not directly available for the entire US or for specific geographic areas of NYS Such information is available from population-based cohort studies that follow people over time in selected communities While the concern underlying this research project is to understand CVD levels in the community ie incidence and prevalence we do not have data that directly provide us with this information In the absence of such data hospitalizations are frequently used as indicators of the burden of disease in communities particularly in surveillance or hypothesis-generating investigations such as this one In NYS the Behavioral Risk Factor Surveillance System (BRFSS) an ongoing monthly telephone survey of adults aged 18 years and older provides information about the burden of CVD by age sex race income and education

CVD Risk Factor Data The BRFSS data for NYS show that over 20 of the population aged 65 and older reported having been diagnosed with CVD versus less than five percent of the population under 65 Males were slightly more likely to report having CVD than females Nonshyhispanic whites compared to other ethnic groups combined were slightly more likely to report having CVD None of these slight differences were statistically significant however The only statistically significant demographic risk factors in the BRFSS data were low income and education levels Statistically significant elevations of reported CVD were observed for respondents with incomes below $25000 and respondents with a high school degree or less education (NYS DOH 2009a)

Another factor that can affect the outcome measure hospitalizations is the likelihood of mortality prior to reaching the hospital Studies show a variety of factors affect whether an individual experiencing a heart attack or stroke outside the hospital survives to be counted in hospitalization data These factors include lack of knowledge about personal risk or prior disease socio-economic status race marital status and distance to the hospital (Galea 2007

4

Barnett 2006 Ayala 2003) Population-based studies report sudden cardiac death usually outside the hospital to be the first overt manifestation of heart disease for 40 to 60 of all cases ( Kannel 1987 deVreede-Swagemakers 1997 UK Heart Attack Study 1998) An analysis of out-of-hospital mortality from AMI in Pennsylvania in 1998 showed that in relatively rural areas where travel time to the hospital was 25 minutes or longer 72 of all AMI deaths occurred outside the hospital compared with a statewide average of 49 of AMI deaths occurring outside of the hospital (OrsquoNeill 2003) In New York State 55 of IHD deaths occur outside the hospital compared to 35 of stroke deaths (NYS DOH 2009b)

Hudson River PCB Contamination

During an approximate 30-year period ending in 1977 the General Electric (GE) plants in Hudson Falls and Fort Edward used PCBs in the manufacture of electrical capacitors and discharged large quantities of PCBs into the upper Hudson River Estimates of the total direct discharges to the river are as high as 1330000 pounds In 1983 the US EPA classified a 200shymile section of the Hudson River from Hudson Falls to New York City as a National Priority List site making it one of the largest Superfund sites in the United States In 2002 the US EPA issued a Record of Decision calling for targeted environmental dredging and removal of PCB-contaminated sediment in the upper River (USEPA 2002) The dredging began in May 2009

The Superfund environmental investigations and planned remedial actions have focused on the areas of the river near Hudson Falls and Fort Edward where the highest levels of PCBs in sediments are located The 2002 Record of Decision provides information about the human health risk assessment conducted for the Upper and Mid-Hudson and the ecological risk assessment conducted for the Upper and Lower Hudson River (USEPA 2002) Fish contamination and fish consumption are the primary concerns at the site

In 1976 fishing was banned in the upper river and in 1995 a catch and release policy replaced the ban NYS DOH continues to recommend people eat no fish from the Upper Hudson River that children under age 15 and women of childbearing age eat no fish from the entire 200 mile stretch of river below Hudson Falls and that the general population eat none of most species of fish caught between the federal dam at Troy and Catskill approximately 40 miles south of Troy The Lower Hudson remains closed to commercial fishing for striped bass and eight other species

The fishing bans and advisories were issued because ingestion of fish is the major potential pathway for exposure for the general population Many previous studies have shown relationships between consumption of fish from contaminated waters and human PCB levels Another potential pathway for exposure less frequently studied to date is inhalation of PCBs NYS DOH and ATSDR launched ldquoThe Hudson River Communities Projectrdquo in 2000 to address this issue (More information on this project is provided below)

POPS and PCBs

The category ldquopersistent organic pollutantsrdquo (POPs) includes man-made chemicals that are by definition persistent in the environment Examples of POPs are polychlorinated biphenyls

5

(PCBs) polychlorinated dibenzo-p-dioxins (PCDDs) polychorinated dibenso-furans (PCDFs) and pprsquo-diphenyldichloroethene (DDE) a breakdown product of the pesticide dichlorodiphenyltrichloroethane (DDT) These chemicals are stored in human fat tissues accumulating over an individualrsquos lifetime They also accumulate in fish and other species thereby concentrating up the food chain Levels of most POPs measured in blood serum of Americans have decreased in the last few decades An exception is polybrominated diphenyl ethers (PBDEs) levels of which have been increasing in recent years While levels of many POPs are declining a large number continue to be detectable in blood for most of the US population (USDHHS 2009)

This follow-up investigation focuses on the Hudson River where the POPs of concern are PCBs PCBs are a group of synthetic chlorinated compounds manufactured in the US through 1977 Production was halted due to concerns about the persistence of PCBs in the environment and the toxicity of PCBs shown in animal models PCBs were used primarily by the electrical industry for capacitors and transformers and were also used in hydraulic fluids fluorescent light fixtures flame retardants inks adhesives carbonless copy paper paints pesticide extenders plasticizers and wire insulators PCBs can be oily liquids or solids and some are volatile and can exist as vapor in air PCBs in the environment are generally mixtures of the many different types known as congeners (ATSDR 2000)

The variety of types of PCBs adds to the complexity of studying adverse animal and human health effects Increasingly PCB research focuses on specific PCB congeners and groups of congeners because specific types of congeners appear to have different mechanisms of action and toxicity Dioxin-like PCBs (DLPCBs) are an important subset of PCBs due to concerns about their toxicity The focus on dioxin-like PCBs is paralleled by increasing research on the group of POPs with dioxin-like properties This type of PCB and POP research assigns toxic potency values relative to dioxin to various PCBs and POPs to provide an overall toxic equivalency value to specific PCB and other POPs

From 2002 through 2007 NYS DOH and SUNY Albany School of Public Health researchers conducted the Hudson River Communities Study among residents of Hudson Falls and Fort Edward the villages where the PCB discharges from two GE electrical capacitor plants occurred The comparison community was upriver from the discharges in Glens Falls This studyrsquos evaluation of exposures included measuring PCBs in outdoor air and in blood serum The study found that PCB levels in outdoor air were statistically significantly higher in the Hudson FallsFort Edward area (072 nanograms per cubic meter) than in the upriver Glens Falls area (040 nanograms per cubic meter) but that the PCB concentrations in the Hudson FallsFort Edward area were lower than those in other communities with known PCB-contaminated sites and similar to levels reported in other locations in the northeastern United States A statistically significant reduction in the total PCB concentrations in outdoor air samples was observed at a distance of 1200 meters (34 mile) from the river from 0760 ngm3 to 0497 ngm3 (Fitzgerald et al 2007 Palmer et al 2008)

The average PCB concentrations in serum for the study and comparison populations did not differ significantly (307 ppb wet weight versus 323 ppb) The average toxic equivalency quantities (TEQ) for dioxin-like PCB congeners for the study and comparison area populations

6

also did not differ In addition the study found no detectable differences in study subjectsrsquo serum PCB levels associated with proximity to contaminated portions of the river or other PCB sources The study did find that lifetime consumption of Hudson River fish was associated with higher serum PCB levels (Fitzgerald 2007 Palmer 2008) Another recent biomonitoring study of Brooklyn Manhattan and New Jersey fisherman who consumed versus those who did not consume their catch from the lower Hudson River and New YorkNew Jersey harbor suggests eating fish from these waters is not associated with an increased body burden of PCBs or most other organochlorines This study did not address the inhalation pathway (Morland 2008)

PCBs and Cardiovascular Disease Outcomes

Research studies on adverse health effects associated with PCB exposure have focused on many types of outcomes organs and biological systems including cancer liver reproductive and developmental effects thyroid endocrine immune system and neurological effects as well as specific effects on the skin and eyes Relatively little research has focused to date on associations between PCB exposures and CVD outcomes The majority of human studies have assessed health effects in occupational groups having relatively high-level exposures among people consuming fish from contaminated waters or people exposed to relatively high levels due to accidental contamination of food (USDHHS NIH 2010)

PCBs have been shown to produce a variety of toxic effects in animals These effects include cancer allergies hypersensitivity damage to the central and peripheral nervous systems reproductive disorders and disruption of the immune system (ATSDR 2000) In contrast research to date does not tell us whether specific types of human health effects have occurred from the low-level exposures experienced by the general population Some human epidemiological studies of relatively low levels of PCDDs PCDFs and PCBs suggest possible relationships not only with neurodevelopment of infants but also diabetes and a condition called insulin resistance conditions that increase CVD risk (Longnecker 2001 Carpenter 2006 Arisawa 2005)

In 1998 a study of a population in Seveso Italy exposed to relatively pure TCDD (dioxin) from an industrial accident showed increased mortality from CVD respiratory diseases and diabetes (Bertazzi 1998) These findings suggested the possibility that other dioxin-like chemicals such as dioxin-like PCBs might increase risk for these diseases Some more recent animal studies described below suggest possible mechanisms of action for PCB effects that increase risk for CVD diabetes and hypertension Another relevant source of information is from research using the biological monitoring data from the National Health and Nutrition Examination Survey (NHANES) project to look at the association between PCB levels in blood and self-reported diabetes and heart disease The body of evidence based on analytical human epidemiological studies remains relatively sparse however

Some recent studies using NHANES biomonitoring data show associations between PCB levels and CVD (Lee 2006 Lee 2007 Ha 2007 Everett 2008) These studies are cross-sectional (comparisons at one point in time) evaluations of PCB levels measured in blood samples from individuals and self-reported CVD and diabetes a risk factor for CVD While these studies use individual-level biomonitoring and self-reported health outcome data they are considered to be

7

relatively weak in terms of providing evidence of causality because both exposure and outcome are measured at the same point in time Given the nature of NHANES data collection and reporting practices these studies are not able to control for possible geographical confounding which may occur if urban versus rural residence is associated with higher PCB levels and higher risk for CVD for example The higher PCB levels among African Americans in these data suggest urban residence may be playing a role in the PCB data

The results of animal studies suggest multiple mechanisms by which PCB exposures could affect cardiovascular risk as well as multiple types of adverse health outcomes One hypothesized mechanisms of action specifically for dioxin-like PCBs is that they may have anti-estrogenic effects which impact CVD risk factors such as serum cholesterol levels and blood pressure (Lind 2004) Several studies suggest PCBs interact with a specific receptor (AhR) and increase cellular oxidative stress causing inflammation cell injury and cardiovascular dysfunction (Hennig 2002) Similarly some studies suggest PCB-induced inflammation of fat cells leads to obesity and increased risk for coronary artery disease (atherosclerosis) (Arsenescu 2008)

3 FOLLOW-UP STUDY DATA SOURCES AND METHODS

This follow-up investigation will provide perspective on the 2004-2005 findings (Sergeev and Carpenter 2005 Shcherbatykh et al 2005) by producing results from analyses conducted at a more detailed level of geography the block group rather than ZIP code in order to better control for confounding When the term confounding is used by epidemioogists it refers to the problem that can arise when there is a factor such as income level that is not the factor of interest in the study but that is related to the risk factor being studied (residential location ndash proxy for exposure) and to the disease being studied (CVD) This type of factor is labeled a ldquoconfounderrdquo because it can confound ie interfere with or confuse the evaluation of associations among the variables of interest in the data All types of epidemiological studies face issues of confounding but ecological ie group-level analyses face additional challenges related to potential confounding because analyses can not account for some important individual differences within the groups that are the units of analysis (Morganstern 1998b Hertz-Picciotto 1998) This investigation was able to control for each individualrsquos age sex and race but not for individual-level risk factors such as income or education level

The follow-up investigationrsquos specific objectives are to

improve the ability to control for confounding by socio-economic status through the use of block groups rather than ZIP codes as the unit of analysis

compare the hospitalization rates among residents of block groups nearest the Hudson River to those among residents of block groups farther from the Hudson River

compare the findings of this follow-up study to those of the earlier studies

Study Area and Exposure Areas

The overall study area is comprised of the 12 counties with geographic areas abutting the Hudson River south of the source of PCB contamination from the General Electric facility in Hudson Falls The 12 counties are Westchester Rockland Putnam Orange Dutchess Ulster

8

Columbia Greene Rensselaer Albany Washington and Saratoga US Census boundaries are used to define the study area because census population estimates provide the data for the study area populationrsquos size and the populationrsquos age sex and race distribution Because the study period spanned 1990 through 2005 and populations change over time 1990 and 2000 Census data were used to linearly interpolate the between-year populations from 1991 through 1999 for each block group 2000 data were used without extrapolation to estimate block group populations from 2001 through 2005

Because block group boundaries change between the Census years 1990 block populations were normalized to fit 2000 block group boundaries before the estimation of sex age and race group specific populations In the digital map 2000 block group boundaries were overlaid on the 1990 block centroids to assign 1990 block population counts to the corresponding 2000 block group boundaries Then 1990 block populations were summed to fit the block group boundaries Next population counts by age group (0-14 15-24 25-34 35-44 45-54 55-64 65-74 75+) sex (male female) and race (white black asian other) were estimated for each year at the block group level This investigation includes all ages over 25 years and the study population is divided into four race categories white black or African-American Asian and ldquoOtherrdquo The Pacific Island group was combined with the Asian group and American Indian ldquootherrdquo and ldquomultiple racesrdquo categories are combined into a larger ldquootherrdquo category

Classification of the population into potential exposure categories for this study is based exclusively on location relative to the Hudson River Two types of exposure proxy categories were developed block group adjacency and block group centroid distance to the river shore For the adjacency categories block groups were assigned to the adjacent category if they directly abutted the Hudson River and all other block groups in the 12 counties were assigned to the nonshyadjacent category

For the distance categories a central point (centroid) within each block group was identified and used to assign block group centroid distance to the river shore The block group centroids were population-weighted which means the geographic distribution of households within 2000 census blocks were used to locate block group centroids based on the geographic population distribution By locating a population-weighted block centroid rather than a more straightforward distance centroid the assignment of block group distance to the river shore more accurately reflects distance to the river for the majority of block group residents Buffers of one-half mile and one mile from the Hudson River boundary were created for the assignment of each block group into one of the three distance categories less than frac12 mile (closest) frac12 to one mile (close) and farther than one mile from the river (distant)

Distance categories were selected a priori at the start of the study Because environmental sampling data were not collected along the length of the river included in this investigation we are not able to determine whether these distance categories represent actual differences in exposures As discussed previously a statistically significant reduction in the total PCB concentrations in outdoor air samples was observed at a distance of 1200 meters (34 mile) from the river from 0760 ngm3 to 0497 ngm3 in a study conducted in the Hudson Falls Fort Edward area near the source of PCBs (Fitzgerald 2007 Palmer 2007)

9

For the adjacency categories of the 2371 block groups in the study area 255 (11) were classified as adjacent to the Hudson River and 2116 (89) were classified as non-adjacent (Figure 2) For the distance categories 225 (9) were classified as closest to the river 216 (9) as close and 1930 block groups (81) as distant from the river

Comparison of Figures 1 and 2 show that block groups include smaller geographic areas than ZIP codes and that block groups are relatively smaller in area in urban versus rural areas The study area includes a total of 368 ZIP codes compared to a total of 2371 block groups ZIP code average population in the study area is approximately 8000 compared to block group average population of 1200 Comparison of Figures 2 and 3 shows that using population-weighted centroid distance to the river produces different exposure areas than adjacency Fifty-three (20) of the adjacent block groups were categorized for the distance indicator as distant because the population in the block group was concentrated in the area farther than one mile from the river On the other hand about 11 of the 2116 non-adjacent block groups were classified as closest (77) or close (162)

Hospitalization Data

Hospitalizations for ischemic heart disease (IHD) and cerebrovascular disease (stroke) were included in this study More specifically hospital discharge records that listed as the principal diagnosis either IHD (ICD-9-CM 410-414 and 4297) or cerebrovascular disease (ICD-9-CM 430-436) and subsets of each of these categories similar to the subsets evaluated in the Sergeev et al and Shcherbatykh studies were included The subsets are AMI and angina subsets of IHD and ischemic and hemorrhagic stroke which are subsets of cerebrovascular disease (stroke) (Principal diagnosis represents the illness for which the person was admitted to the hospital) Angina is an outcome not evaluated in the prior studies It was added here for comparative purposes because it represents an outcome appropriately addressed by primary outpatient rather than inpatient hospital care Angina is one of a group of outcomes included in the recently developed national ldquoPrevention Quality Indicatorsrdquo (USDHHS AHRQ 2007) as an indicator for geographic areas or populations for whom primary care is inadequate Table 3 lists the specific ICD-9-CM codes and definitions included in this investigation

While one of the previously published studies (Huang et al 2006) showed elevations in hypertension (high blood pressure) rates this outcome is rarely listed as the principal reason for hospitalization As a secondary diagnosis however hypertension is extremely common with substantial overlap with the primary diagnoses of IHD and stroke In addition hypertension hospitalizations were not as elevated in the ZIP codes adjacent to the Hudson River as the IHD and stroke outcomes As a practical matter geocoding the hypertension hospitalization data (1364000 hospitalizations) would overwhelm the resources for this project For these reasons hypertension was not included in this follow-up investigation

The source of the hospitalization data was the NYS DOH Statewide Planning and Research Cooperative System (SPARCS) established in 1979 to collect detailed records on discharges from hospitals located in New York State Persons seen in the emergency department but not admitted are not included in the SPARCS data Hospitalizations at US Veterans Administration

10

Hospitals are also not included in this dataset We identified the hospital transfers in the data and excluded them so that individuals are only counted once for the same event

The NYS DOH Data Protection Review Board for SPARCS hospitalization data approved our use of hospitalization records for this investigation and provided approximately 460000 records with a principal diagnosis of either IHD or stroke Because of the large number of cardiovascular hospitalizations and this studyrsquos requirement for geocoding each hospitalized personrsquos address we selected a 50 random sample of the hospital discharges with principal diagnoses of IHD or stroke for inclusion in the study (229948 records) Approximately one percent of these records (2737) were eliminated from the study because they contained no residential address resulting in 227211 hospitalization records 24939 of these records were eliminated after geocoding because they were not within the 12 county study area This left 202272 records remaining in the study

For geocoding to the block group level the hospitalization street addresses were modified as needed using US Postal Service standards The addresses were assigned geographic coordinates using commercially available geocoding software (MapMarker 2004) A combination of land parcel data and street centerline files were used to assign geographic coordinates No contact was made with cases parents legal guardians or next of kin of cases to determine residential locations

Once geographic coordinates were assigned to cases through address-matching the case locations were overlaid onto digital maps of the study area using a geographic information system so that the number of observed cases falling within the study area boundaries could be determined and a 2000 Census block group could be assigned to each hospitalization In order to protect confidentiality no maps of individual case locations are provided Geographic coordinates were not able to be assigned for some addresses for example rural routes or post office boxes Internet searches were conducted for addresses that were located in facilities such as nursing homes and trailer parks

While fewer than one percent of records were missing sex or age information approximately 15000 records seven percent were missing race information Evaluation of the geographic and temporal distribution of these records showed they were not randomly geographically distributed 47 of the records with no race information were from one hospital for that hospital race was missing from 100 of the records in all years except one after 1997 mid-way through the study time-frame Evaluation of other records missing race showed similar patterns of 100 of records missing race in specific years In order to prevent biased hospitalization rate estimates records missing race were assigned race randomly assuming a race distribution equal to that of the individualrsquos residential block group

Statistical Analysis

This investigation used negative binomial multivariable regression models the statistical methods used in the previous studies (Sergeev and Carpenter 2005 [IHD] Shcherbatykh et al 2005 [stroke]) The follow-up studyrsquos analyses compared hospitalization rates for populations in block groups adjacent or in close proximity to the Hudson River to rates for populations more

11

distant from the Hudson River while adjusting for the potential influence of other factors Sex age and race-specific hospitalization count data comprise the dependent variable and the statistical methods attempt to predict the counts using the sex age and race distributions from Census data as well as block group-level demographics such as median income Models were first estimated using basic Census information for age sex and race Then more fully adjusted models were estimated using information at the block group level for income education population density Hispanic ethnicity and distance to the nearest hospital

More specifically each hospitalized personrsquos block group was used to assign quartile categories of block group median household income of population with less than a college education population density of the population identified as Hispanic and distance from the block group centroid to the nearest hospital These variables were selected in advance The population density variable was selected to account for potential effects on hospitalization rates of large differences in population density from the relatively rural northern counties to the more urban southern counties in the study area of the population identified as Hispanic was included because Hispanic ethnicity may be associated with CVD risk but Hispanic ethnicity was not available in the hospitalization records Distance to the hospital was included because studies show that shorter distances increase the likelihood that individuals with AMI and stroke survive to be admitted to the hospital

As described previously the literature about risk factors for CVD shows strong associations between lower income and education and higher CVD risk Regarding the effects of income and education on hospitalization rates the literature from analytic studies specifically of hospitalizations for CVD is quite sparse Most analytic studies focus on prevalence incidence or mortality in population-based studies or cohort studies and there are not many of these An example of a study that was useful as background is a study of income level and CVD mortality that showed strong associations of lower income and higher CVD mortality for men and women with the relationship for men being linear throughout the range of income levels while for women the relationship was only strong at levels of income below the median (Rehkopf 2008)

This follow-up studyrsquos approach was to use the same set of variables for the more fully adjusted models using quartile cut points for all the sexrace groups and types of CVD diagnostic subsets The 2000 Census data were used to create the population-weighted quartile categories to be used in the statistical analyses The quartile cut points are shown below

population density(persons per square mile)

median 1999 household income

with less than a college education

with Hispanic ethnicity

Hospital distance

Quartile 1 61 - lt507 $2499-lt$40968 0 - lt31 0 - lt2 lt25 (median) Quartile 2 507 - lt102 $40968 - lt54650 31 - lt43 2 - lt5 25 - lt6 (50-75) Quartile 3 102 - lt6410 $54650 - lt72944 43 - lt55 5 - lt10 6 ndash 11 (75-90) Quartile 4 6410 - 94372 $72944 - 200001 55 - 100 10 - 86 11 ndash 45 (90-100)

SAS version 9 provided the statistical programs used to estimate hospitalization rates for exposure proxy categories and demographic categories The models provide ldquocontrast

12

estimatesrdquo ie rate ratios for one category versus a reference category of a variable and 95 confidence intervals (CI) for these rate ratios For the location categories rate ratios for the areas adjacent are compared to areas not adjacent to the river (reference) and areas closest and close to the river are compared to areas distant from the river (reference) If a rate ratio is greater than 100 the model has estimated an elevation of hospitalizations in that category of the study population compared to the reference population If the rate ratio is less than 100 then there are fewer than expected hospitalizations compared to the reference population

The magnitude of the elevation or deficit is also estimated by the rate ratio For instance if the hospitalization rate is twice as high in the adjacent population it would result in a rate ratio of 200 while a 50 higher rate would result in a rate ratio of 150 If the hospitalization rate in the comparison group was only half as big as in the reference group this would result in a rate ratio of 050 The 95 confidence interval assists with determining whether the observed differences are statistically meaningful ie statistically significant The confidence interval provides the range in which there is a 95 probability of including the true rate ratio

In addition to directly assess potential confounding we examined the associations between income and exposure proxy categories and between income and CVD hospitalization risk Because the modeling process provides rate ratios that provide comparisons among groups but not the rates themselves we also calculated the rates per population for each sex race and income category in order to directly observe the relative magnitudes of CVD hospitalization rates Finally analyses stratified by income quartile were also conducted in order to assess the potential role of confounding by income The stratified analyses consist of negative binomial regressions conducted separately for each quartile of income

4 RESULTS

Demographics Associations with Proximity to the River

Study Population Table 4 provides detailed demographic characteristics for the total study population and the populations of the adjacent versus non-adjacent and nearer versus farther from the river block groups from the 2000 Census The total population of the study area for this one year of the study time period is 2924631 persons The first row of Table 4 shows that approximately 11 percent of the study area population is classified as living in block groups adjacent to the Hudson River and 89 live in non-adjacent block groups Approximately 8 of the studyrsquos population lives in the closest block groups approximately 10 lives in close block groups and approximately 82 in distant block groups The table also shows that approximately 81 of the total study population is identified as white 9 black 3 Asian less than 01 Pacific Islander 02 American Indian 4 ldquootherrdquo and 2 ldquomultiple racesrdquo

The 2000 Census data (Table 4) also show that the populations in block groups adjacent and nonshyadjacent to the river are similar in terms of the percent of minority population but the adjacent block groups include a smaller percentage of persons who have attended college and a higher percentage of households below the poverty level Median income is $47320 for block groups adjacent versus $62668 for block groups not adjacent to the river Data for the study area from the 1990 Census show similar distributions (data not shown)

13

For the distance to the river categories there is a gradient for the three levels of distance from the river with the closest block groups on average having higher percentages of minorities smaller percentages of persons who have attended college and higher percentages of households below the poverty level (Table 4) The populations of the close and closest block groups are more similar to each other than to the populations of the distant block groups The population of the closest block groups is estimated to be 19 black close block groups 14 black and distant block groups 8 black The proportions for the ldquootherrdquo category used in the analyses which includes multiple races and American Indians are approximately 12 in both the closest block groups and in the close block groups compared to 5 in the block groups most distant from the river Median income for the closest block groups is $40748 close block groups $44962 and most distant block groups $64847

The bar chart below illustrates the correlations between US Census 2000 indicators of block group socio-economic status and block group centroid (population weighted) distance from the river for the study population

2000 Census Demographics for Study Population by Distance to River Category

black other hispanic no college below poverty

of pop ulati on

0

10

20

30

40

50

60

lt12 mile

12 to 1 mile

gt1 mile

Table 5 shows selected demographic information for the person-year estimates used in the study analyses for the 1990-2005 study period Person-year data represent individuals from Census estimates counted repeatedly for each year of the study timeframe As described in the Methods section the estimation of person years used interpolation to account for population changes from 1990-2000 and techniques for taking account of block boundary changes The person-years estimates for demographic categories (complete data not shown) show similar patterns as the 1990 Census data (data not shown) and 2000 Census data (Table 5) Table 5 for example shows nearly identical percentages of population in each of the river distance categories as Table 4

Table 5 also shows person year information for income quartiles and exposure proxy categories as well as race and exposure proxy categories that is consistent with the Table 4 data for the

14

2000 population Table 5 row percentages show for example that while approximately 11 of the total study population live in adjacent block groups approximately 19 of the population in the lowest income quartile live in adjacent block groups By design approximately 25 of the total study population is in each of the income quartiles Table 5 column percentages show however that in the adjacent block groups 41 of the population is in the lowest income quartile and only 9 are in the highest income quartile

For the block groups within frac12 mile of the river the distribution among income quartiles is even more skewed towards the lowest income quartile 61 of the population living within frac12 mile of the river is in the lowest income quartile and only 6 is in the highest income quartile Given that approximately 25 of the total study population is within each income quartile these percentages indicate that living within the block groups closest to the river is associated with more than a doubling of the likelihood of being in a block group in the lowest socio-economic status quartile 436 of those living in the block groups close to the river (12 to one mile from the river) are in block groups in the lowest income quartile While less than a doubling this indicates that living in the close block groups is associated with a 75 increase in the likelihood of being in the lowest quartile for socio-economic status

Hospitalizations Table 6 shows the process of removing hospitalization records from the study for a variety of reasons The first column includes the 202272 records that contained a residential address and were located within the 12-county study area as described in the Methods section The next five columns show the records remaining after data cleaning exclusion of transfers elimination of records with missing covariates and exclusions due to poor geocoding results Approximately 33000 records were eliminated because they were transfers from one hospital to another for the same event For approximately 1100 records geocoding was inadequate for assigning block group for imputing race as described in the Methods section 336 records were missing sex (11) or age (325) Of the 167733 records remaining 154220 (92) were able to be adequately geocoded to a street address for assignment in a specific block group While approximately eight percent of hospitalizations were eliminated due to inadequate geocoding Table 6 shows that the percentages of hospitalization records for white black Asian and ldquootherrdquo categories and the distribution among diagnostic categories change very little after these records were dropped

The final column of Table 6 shows that of the 154220 hospitalizations in the analysis 101012 IHD hospitalizations comprise the largest health outcome category and are 65 of the hospitalizations in the study AMI hospitalizations (35811) account for 35 of the IHD hospitalizations and angina (4731) accounts for five percent of the IHD hospitalizations Stroke hospitalizations (53208) comprise 34 of the studyrsquos hospitalizations Approximately 36 of the stroke hospitalizations are for ischemic stroke (19371) and approximately 11 are for hemorrhagic stroke (5995)

Table 7 shows the distributions of hospitalizations among the four income quartiles While each income quartile (by definition) contains approximately 25 of the study population for all types of hospitalizations we see that the lowest income quartile includes more than 25 of hospitalizations from a minimum of 286 (IHD) to a maximum of 356 of hospitalizations (angina) The highest income quartile includes fewer than 25 of hospitalizations for each CVD

15

category ranging from 182 for angina to 224 for hemorrhagic stroke These deviations from the 25 quartile distribution show a strong linear pattern with the largest percentage of hospitalizations coming from residents of block groups in the lowest income quartile and the smallest percentages coming from residents of the most wealthy block groups For AMI the lowest income quartilersquos share of hospitalizations is 50 higher than the highest income quartilersquos share of hospitalizations (30 vs 20) IHD and stroke show similar differences and angina shows a doubling with the lowest income quartile contributing 36 of hospitalizations compared to the highest income quartilersquos 18 share of hospitalizations

Table 7 also shows the distribution of hospitalizations among the race categories and the study population race category distribution for comparison Blacks show lower percentages of IHD and AMI than their population percentages and higher percentages of angina and stroke While full presentation of race and sex findings is beyond the scope of this report the lower than expected percentages of IHD and AMI among blacks who on average have lower socioshyeconomic status and are at greater risk for CVD hospitalizations raised questions that were addressed by calculating hospitalization rates per 1000 population for each sex and race group for each income quartile (Appendix B Table 2) Figures 4 5 and 6 show these rates for the white black and Asian categories (These rates for the ldquootherrdquo category are comparatively very high and are not shown on these graphs This issue is addressed in the Discussion section) Figures 4 5 and 6 show the consistent and strong effects of income quartile on hospitalization rates for AMI and stroke for whites blacks and Asians of both sexes The other outcomes show similar patterns (The ldquootherrdquo group showed an opposite pattern [See Discussion Section])

Demographics Summary

The strong associations of lower income with residence closer to the river (Table 5) and lower income with higher hospitalization rates (Tables 7 amp 8) create a difficult challenge for this investigation The multivariable analyses of hospitalization rates attempt to control for socioshyeconomic differences However because these income differences which strongly affect hospitalization rates are also strongly correlated with the gradations of exposure (river distance) it will be difficult to disentangle the influence of socio-economic status from the influence of potential exposures

Statistical Analyses Proximity to the River and CVD Hospitalization Risk

All the variables described above were included in the multivariable regression models and almost every quartile category for every variable showed statistically significant and expected types of effects for all the CVD hospitalization categories (Appendix B Tables 1a and 1b) The following presentation of results focuses on the rate ratios for the exposure proxy variables Table 8a (IHD) and 8b (stroke) show the regression modeling results Results are shown first for models that include only age sex and race in addition to the exposure proxy categories and second for models that also include median income education population density Hispanic ethnicity and distance to nearest hospital The tables provide numbers of hospitalizations and person-years to assist with interpretation For example for total IHD we can see that while 108 of the study population lives in adjacent block groups 122 of the hospitalizations are among people in adjacent block groups This difference of 14 percentage points is the basis for

16

the finding of a statistically significantly elevated rate ratio (risk) for IHD among people living in the adjacent block groups (RR=133 CI 112-157)

Looking across the columns of Tables 8a and 8b we can see the rate ratios for each of the outcomes for the adjacent versus non-adjacent block groups and the results are similar to those for IHD The elevated rate ratios range from 123 for angina to 136 for AMI The only rate ratio not elevated in these first models adjusting only for age sex and race is for hemorrhagic stroke Going down the rows of the Table the next set of rate ratios are for the distance to the river categories The findings are very similar to those for the adjacent block groups Here however we have two exposure proxy categories allowing comparison of the two These rate ratios show no evidence that living in the block groups closest to the river (lt12 mile to the river) produces greater risk than living in the block groups close to the river (12 to 1 mile from the river)

The next set of rate ratios are for models with additional adjustment for the demographics of each hospitalized personrsquos block group as well as distance to the nearest hospital As expected adjustment for block group median income and other demographics accounts for some of the variability in hospitalization counts and thereby reduces the magnitude of the rate ratios estimated for the river adjacency and distance categories The rate ratio elevation for total IHD for example is reduced from 133 to 107 (RR=107 CI 104-111) The elevated rate ratios in these more fully adjusted models for the adjacency analyses are of similarly small magnitude ranging from 106 for AMI to 110 for ischemic stroke Four of the six outcomes show statistically significant elevations Again only hemorrhagic stroke shows no evidence of an elevated rate ratio

For the river distance categories none of the rate ratios show any evidence that living in the block groups closest to the river (lt12 mile to the river) produces greater risk than living in the block groups close to the river (12 to 1 mile) Of the six types of outcomes only two show statistically significantly elevated rate ratios for living in the closest block groups (total stroke and ischemic stroke) For the close block groups four out of six outcomes show statistically significantly elevated rate ratios The statistically significantly elevated rate ratios vary from 106 (AMI stroke) to 113 (angina)

To address the strong evidence described earlier for confounding in these data ie the association of lower income with both CVD hospitalizations and residence closer to the river we also conducted the analyses separately for each quartile of income These results are presented in Table 9 The first results are from analyses adjusting only for age sex and race and the second set of results are from models including the additional demographic factors For simplicity rate ratios are shown only for block groups closest to the river compared as in the other models to block groups farther than one mile from the river (Results were similar for the other distance categories (data not shown)

For the two lowest income quartiles Table 9 shows results similar to those for the study population as a whole (Table 8a 8b) However for the two highest income quartiles which represent frac12 of the study population Table 9 shows a consistent pattern of reduced rate ratios for residence in the closest block groups within frac12 mile of the river In the highest income quartile

17

EXECUTIVE SUMMARY

Background In October 2006 Member of Congress Maurice Hinchey requested that ATSDR conduct a health consultation to assess cardiovascular disease among people residing near the Hudson River A 200-mile segment of the Hudson River from Hudson Falls south to New York City is a National Priorities List (NPL) (Superfund) site due to contamination of the river with polychlorinated biphenyls (PCBs) from GE sites in Hudson Falls and Ft Edward In support of his request Member of Congress Hinchey cited three studies of cardiovascular hospitalizations conducted with similar data and methods These 2004-2005 statewide studies suggest residents of ZIP codes containing hazardous waste sites with persistent organic pollutants (POPs) such as PCBs have elevated risk for hospitalizations for cardiovascular disease including ischemic heart disease stroke and hypertension Analyses conducted separately for ZIP codes next to the Hudson River suggest residents of these specific ZIP codes also have increased rates of hospitalization for cardiovascular disease

In response to Congressman Hincheyrsquos request for follow-up of the 2004-2005 findings for the Hudson River area ATSDR and NYS DOH agreed to conduct a similar group-level study but at a different level of geography the Census block group By using a level of geography that is generally smaller in area and population than a ZIP code this follow-up seeks to more accurately adjust for socio-economic and other factors that affect cardiovascular hospitalizations and to improve upon the geographic classification of relative distance to the Hudson River The follow-up investigationrsquos analyses are restricted to the twelve counties abutting the Hudson River from Washington County south to Westchester County Within the study arearsquos twelve counties cardiovascular hospitalization rates in areas closest to the Hudson River are compared with rates in areas farther from the river

Objectives The follow-up investigationrsquos specific objectives are to increase the studyrsquos ability to control for confounding by socio-economic status through the

use of block groups rather than ZIP codes as the unit of analysis compare hospitalization rates among residents of block groups nearest the Hudson River to

those among residents of block groups farther from the Hudson River and compare the findings of this review to those of the earlier studies

Methods This study examines CVD hospitalization rates and residence near the Hudson River in the 12 counties abutting the Hudson River below Hudson Falls from 1990 through 2005 The hospitalization data for ischemic heart disease and stroke were acquired from the NYS DOH Statewide Planning and Research Cooperative System (SPARCS) established in 1979 to collect detailed records on discharges from hospitals located in New York State Statewide US Census data at the block group level from 1990 and 2000 were used to calculate the number of people in each race sex and age group for the study years

Classification of the population into potential exposure categories was based exclusively on location relative to the Hudson River Two types of exposure proxy categories were developed block group adjacency and block group centroid distance to the river shore For the adjacency categories the hospitalizations were assigned to the adjacent category if the residential address was in a block group directly abutting the Hudson River For the distance categories block

ii

group centroid (population-weighted) distance to the river shore was used to assign hospitalizations in each block group to one of three categories less than one-half mile from the Hudson River (closest) between one-half to one mile (close) from the river and farther than one mile from the river (distant) Multivariable regression analyses were conducted to compare rates of hospitalizations among the exposure categories assigned at the block group level These analyses controlled for age sex and race using individual-level data and population density percent Hispanic median income education and distance to the nearest hospital at the block group level

Results The use of block groups resulted in 2371 block groups as the units of analysis for the study rather than the 368 ZIP codes in the 12-County study area The average population per block group in the study area is 1200 compared to average population of 8000 per ZIP code The smaller geographical area and smaller populations in block groups contributed to improving the accuracy of the assignment of income and other group-level demographics to the hospitalizations included in the study and enhancing control for confounding by socio-economic status (Tables 8a and 8b)

The studyrsquos multivariable analyses showed relatively small but statistically significant elevations of CVD hospitalization risk for residents of block groups categorized as adjacent in close and closest proximity to the Hudson River There is no evidence that residents of the closest block groups (less than frac12 mile) had higher hospitalization rates than residents of the close block groups (between frac12 to 1 mile) If there were a strong association between exposures from living near the river and the risk of CVD hospitalization such a ldquodose-responserdquo might have been shown by the analyses

The data showed strong statistically significant associations between CVD risk and socioshyeconomic status To enhance the control for socio-economic status separate analyses for each income quartile were conducted These stratified analyses showed that residence in block groups near (close closest or adjacent to) the Hudson River was associated with increased risk for CVD hospitalizations among residents of lower income block groups Conversely residence in block groups near the Hudson River was associated with decreased risk for CVD hospitalizations among residents of higher income block groups In other words in analyses that held income constant by looking only within the lower or higher income populations living near the river shows opposite effects depending upon which income group is considered The estimated increased and decreased risks are of sufficient magnitude to cancel each other out when average rate ratios for the four income quartiles are calculated The average rate ratios can be interpreted as showing no evidence that hospitalizations among the general population are elevated for residents living near the river

This type of group-level statistical analysis is not able to precisely identify the reason for these differing results However the strength and direction of associations among variables in the statistical models can be evaluated in light of known risk factors to suggest likely explanations In this case socio-economic status appears to be the most likely explanation for the findings The strong association between socio-economic status and residential location (the indicator of potential exposure) and between socio-economic status and CVD hospitalization risk makes it likely that the small but statistically significant elevations shown in the multivariate models

iii

(Tables 8a amp 8b) are the result of the inability to completely control for socio-economic statusrsquo effects on CVD hospitalization risk The strong effect of socio-economic status on residential location (exposure proxy) and on CVD hospitalization risk makes it likely that these overall multivariable regression results are affected by uncontrolled ie residual confounding by socioshyeconomic status

Because of the similarities in study design and statistical analysis the prior studyrsquos findings can be compared to some degree with the findings of the multivariate regression analyses in the current study The major difference between the studies is the composition of the comparison areas for the analyses Because of this difference as well as other differences described in the report the studies are not entirely comparable

CVD hospitalization rate ratios estimated for living in block groups adjacent closest or close to the river in the current study are substantially lower than the rate ratios for residence in ZIP codes adjacent to the Hudson River in the previously published studies In most cases the current analyses show elevations below 10 in contrast to elevations of 20 to 39 in the prior studies The results of the current study are quite similar to the prior study results in terms of the associations reported for age sex race and income These similarities and the general methodological similarities between the current and prior studies suggest the likelihood that the prior studiesrsquo findings also were affected by confounding by socio-economic status

Conclusions and Recommendations While this type of study can neither prove nor disprove causal links among risk factors and health effects it is useful for examining evidence for unusual patterns that might warrant additional investigation This evaluation of hospitalization data residential location and socio-economic factors in block groups in the 12 Counties abutting the Hudson River revealed a closely coinciding pattern of lower socio-economic status and residential proximity to the Hudson River the proxy for potential exposure and a strong association of lower socio-economic status and risk for CVD hospitalizations The strong associations among residential location socio-economic status and CVD risk prevent us from drawing definitive conclusions from the multivariable regression analyses that showed modest but statistically significant elevations of CVD hospitalization risk for residents of block groups near the river

To draw more definitive conclusions particularly for outcomes such as CVD that are determined by multiple known risk factors studies seeking to investigate the role of PCB or POP exposures will require resource-intensive methods that include gathering individual-level exposure or biomonitoring information as well as medical histories and information on other risk factors

This current investigation did not include environmental or biological sampling data One study did gather such data in the Hudson River region with the highest potential exposures (Hudson Falls Fort Edward) This study detected a statistically significant reduction in the total PCB concentrations in outdoor air samples at a distance of 1200 meters (34 mile) from the river However the PCB levels in outdoor air in this area were lower than those in other communities with known PCB-contaminated sites and similar to levels reported in other locations in the northeastern United States The study found no detectable differences in study subjectsrsquo serum

iv

PCB levels associated with proximity to contaminated portions of the river or other PCB sources (Fitzgerald 2007 Palmer 2008)

It is recommended that the results of the current investigation of CVD hospitalization rates and residence near the Hudson River be shared with NYS residents and other stakeholders with interest in this issue and be submitted for publication in a peer-reviewed journal

v

1 INTRODUCTION

The New York State Department of Health (NYS DOH) has a cooperative agreement with the Agency for Toxic Substances and Disease Registry (ATSDR) to perform environmental exposure and health assessments conduct health statistics reviews and perform epidemiological studies of populations in New York State which may have been exposed to environmental contaminants In October 2006 Member of Congress Maurice Hinchey requested that ATSDR conduct a health consultation to assess cardiovascular disease among people residing near the Hudson River A 200-mile segment of the Hudson River from Hudson Falls south to New York City is a National Priority List (NPL) (Superfund) site due to contamination of the river with polychlorinated biphenyls (PCBs) from GE sites in Hudson Falls and Ft Edward (See Appendix A for Member of Congress Hincheyrsquos letter)

In support of his request Member of Congress Hinchey cited three studies (Sergeev and Carpenter 2005 [ischemic heart disease] Shcherbatykh et al 2005 [stroke] Huang et al 2006 [hypertension]) These statewide studies suggest residents of ZIP codes containing hazardous waste sites with persistent organic pollutants (POPs) such as PCBs have elevated risk for hospitalizations for cardiovascular disease including ischemic heart disease stroke and hypertension Analyses conducted separately for ZIP codes next to the Hudson River suggest residents of these specific ZIP codes also have increased rates of hospitalization for cardiovascular disease

These studies published in 2005-2006 are ecological studies In an ecological study data are analyzed by evaluating risk factors and health outcomes among groups For example rates of disease are compared for populations of ZIP codes with varying median income levels to see if income levels appear to be related to disease outcome rates In these previously conducted studies rates of hospitalization for residents of ZIP codes with hazardous waste sites containing POPs were compared with hospitalization rates for residents of ZIP codes without such waste sites to look for evidence that potential exposures might be impacting health outcomes The studies report statistically significant elevations of cardiovascular disease hospitalizations in ZIP codes with POPs For coronary heart disease and a subset acute myocardial infarction the study showed elevations of approximately 15 and 20 respectively (Sergeev and Carpenter 2005) for stroke a 15 elevation (Shcherbatykh et al 2005) and for hypertension a 19 elevation (Huang et al 2006) in ZIP codes containing waste sites with POPs compared with hospitalization rates in other New York State ZIP codes (excluding NYC) that did not contain waste sites with POPs (Table 1)

In sub-analyses in these studies cardiovascular hospitalization rates for ZIP codes adjacent to the Hudson River were compared with cardiovascular hospitalization rates for ZIP codes in the rest of the state (excluding NYC) identified as not having inactive hazardous waste sites with POPs These sub-analyses showed statistically significantly elevated hospitalization rates of 36 and 39 for coronary heart disease and the subset acute myocardial infarction a 20 elevation for stroke and a 14 elevation for hypertension (Table 1) The Sergeev and Carpenter Shcherbatykh et al and Huang et al articles state that the Hudson River ZIP codes are in an area where people have higher income smoke less exercise more and have healthier diets This information is cited in these articles as evidence that known risk factors related to lower socioshy

1

economic status are likely not the cause for the elevated hospitalization rates in these Hudson River area ZIP codes

In these previously published reports Behavioral Risk Factor Surveillance System (BRFSS) data for entire counties are provided to support statements about the population of Hudson River ZIP codes having higher income and healthier lifestyles However because the ZIP codes adjacent to the Hudson River include only a proportion of the population of the counties abutting the river these county-level BRFSS data may be misleading regarding characteristics of the ZIP codes adjacent to the river (Figure 1) In fact data from the US Census show that the ZIP codes adjacent to the river have lower median income than the counties within which they are located The median income for the total population of the twelve counties adjacent to the Hudson River was $61010 in 2000 while the median income for the ZIP codes adjacent to the river was $45356 This median income for the population of the ZIP codes adjacent to the river is also slightly lower than the statewide (excluding NYC) median income of $47735 (US Census 2000) Thus the Hudson River ZIP codes analyzed in the Sergeev and Carpenter Shcherbatykh et al and Huang et al studies include a population of lower not higher socio-economic status compared to the rest of NYS (excluding NYC)

Because CVD and many CVD risk factors are associated with lower socio-economic status it is especially important to adjust as much as possible for socio-economic status when conducting comparative analyses of CVD hospitalizations or disease incidence or prevalence Regarding the hypothesis about POPs and CVD research studies described in more detail below have suggested possible biological mechanisms for POP exposuresrsquo effects on the heart There is very little research on human exposures to POPs and potential cardiovascular effects however

In response to Congressman Hincheyrsquos request for follow-up of the 2004-2005 findings ATSDR and NYS DOH agreed to conduct a similar study also ecological but at a different level of geography the Census block group By using a level of geography that is generally smaller in area and population than a ZIP code this follow-up seeks to more accurately adjust for socioshyeconomic and other factors that affect cardiovascular hospitalizations and to improve upon the geographic classification of relative distance to the Hudson River In addition rather than comparing hospitalizations among residents of Hudson River area ZIP codes to hospitalizations among residents of all other NYS (exclusive of NYC) ZIP codes with no waste sites with POPs this follow-up investigationrsquos analyses are restricted to the twelve counties abutting the Hudson river from Washington County south to Westchester County Within the study arearsquos twelve counties cardiovascular hospitalization rates in areas closest to the Hudson River are compared with rates in areas farther from the river This more localized approach may also assist with reducing the potential impact of regional differences in medical care practice or reporting factors that could affect the hospitalization data (Ko et al 2007) (Table 2 shows a more detailed listing of the methods used in the prior studies and follow-up study)

Regarding environmental exposures it is important to note that living near a hazardous waste site that contains PCBs or other POPs does not necessarily mean that human exposure from the site has occurred Depending upon the conditions at a site or the specific chemicals exposures may be highly improbable Information on body burden for example PCB levels in blood or on levels of the chemical in off-site air dust or soil is needed to support a claim of exposure The

2

studies published in 2005-2006 did not provide specific information such as environmental or biomonitoring data in support of the assumption that residence in ZIP codes containing hazardous waste sites with POPs or residence in ZIP codes adjacent to the Hudson River resulted in increased exposures to POPs in the population For the purpose of this follow-up we did not conduct environmental or human biological sampling to address this issue Rather this follow-up replicates the prior studiesrsquo approach with some enhancements to better control for confounding particularly confounding by socio-economic status

This descriptive epidemiologic investigation uses already existing hospitalization data from the Statewide Planning and Research Cooperative System and population data from the United States Census to evaluate comparative rates of cardiovascular hospitalizations While this type of investigation can not show proof of cause and effect it can assist in interpreting relationships among indicators of CVD burden and hypothesized exposures associated with distance from the Hudson River This descriptive investigation may also assist in determining whether additional health investigations of these associations are warranted

2 BACKGROUND

Cardiovascular Disease

The widely known risk factors for cardiovascular disease (CVD) in addition to age sex race and family history include tobacco use high blood pressure high serum cholesterol alcohol use obesity low fruit and vegetable intake physical inactivity diabetes low socio-economic status mental ill-health psychosocial stress and the use of certain medications (WHO 2010 NYS DOH 2009 Wing 1988 NYS DOH 2009a)

CVD Definitions CVD is a general term for any disease of the circulatory system The two main types of CVD are ischemic heart disease also termed coronary heart disease and cerebrovascular disease also called stroke Ischemic heart disease (IHD) results from restricted blood flow through the arteries supplying the heart muscle The most serious risk from IHD is acute myocardial infarction (AMI) another term for heart attack AMI occurs when blood flow to the heart is suddenly reduced or stopped usually due to a blockage in a coronary artery Angina is severe constricting pain in the chest often radiating from the area immediately over the heart to the left shoulder and down the arm Cerebrovascular disease refers primarily to stroke the sudden impairment of brain function resulting from interruption of circulation to the brain Ischemic stroke results from blockage of an artery supplying blood to the brain Hemorrhagic stroke occurs due to the escape of blood from a ruptured artery supplying blood to the brain

CVD Burden and Trends CVD is the leading cause of death for men and women of all ages in the United States and in New York State In 1999 for every death from IHD there were five hospitalizations for ischemic disease in NYS (NYS DOH 2009a) Mortality rates from CVD have been declining steadily since the 1980rsquos Research studies estimate that approximately half of the decline is due to improved treatments for CVD and its risk factors such as high cholesterol and high blood pressure and approximately half is due to decreasing incidence of CVD The decrease in incidence is due to trends in health behaviors primarily reduced smoking rates

3

CVD mortality rates do not appear to be declining for all segments of the population however and may actually be increasing among some groups particularly people of lower socio-economic status (Barnett et al 1999) This may be due in part to higher rates of smoking among people of lower socio-economic status as well as to other factors related to socio-economic status and race A growing literature points to multiple causes for CVD and other health disparities including health practices psychosocial stress limited resources discrimination and access to health care (Health United States 2007) Increases in the prevalence of diabetes and obesity in the general population partially offset advances in treatment and favorable trends for other risk factors An increasing ischemic heart disease mortality rate in women ages 35 to 54 since 1997 is attributed to these unfavorable risk factor trends (Ford 2007 Lloyd-Jones 2009) New York State (NYS) and other northeast states have high IHD and low stroke mortality with NYS having the highest IHD mortality and lowest stroke mortality among the 50 states (Howard et al 2009)

CVD Data Sources CVD mortality rates are more easily measured than incidence prevalence or hospitalizations because mortality data from vital records are a comprehensive and relatively accessible data source Information for estimating the incidence and prevalence of CVD outcomes comes primarily from health interview physical examination and record surveys Hospitalization data are available for NYS and the US but are generally collected and analyzed in order to track trends in hospital care and resources expended Actual CVD incidence data validated by physician diagnosis physician examination or reviews of medical records are not directly available for the entire US or for specific geographic areas of NYS Such information is available from population-based cohort studies that follow people over time in selected communities While the concern underlying this research project is to understand CVD levels in the community ie incidence and prevalence we do not have data that directly provide us with this information In the absence of such data hospitalizations are frequently used as indicators of the burden of disease in communities particularly in surveillance or hypothesis-generating investigations such as this one In NYS the Behavioral Risk Factor Surveillance System (BRFSS) an ongoing monthly telephone survey of adults aged 18 years and older provides information about the burden of CVD by age sex race income and education

CVD Risk Factor Data The BRFSS data for NYS show that over 20 of the population aged 65 and older reported having been diagnosed with CVD versus less than five percent of the population under 65 Males were slightly more likely to report having CVD than females Nonshyhispanic whites compared to other ethnic groups combined were slightly more likely to report having CVD None of these slight differences were statistically significant however The only statistically significant demographic risk factors in the BRFSS data were low income and education levels Statistically significant elevations of reported CVD were observed for respondents with incomes below $25000 and respondents with a high school degree or less education (NYS DOH 2009a)

Another factor that can affect the outcome measure hospitalizations is the likelihood of mortality prior to reaching the hospital Studies show a variety of factors affect whether an individual experiencing a heart attack or stroke outside the hospital survives to be counted in hospitalization data These factors include lack of knowledge about personal risk or prior disease socio-economic status race marital status and distance to the hospital (Galea 2007

4

Barnett 2006 Ayala 2003) Population-based studies report sudden cardiac death usually outside the hospital to be the first overt manifestation of heart disease for 40 to 60 of all cases ( Kannel 1987 deVreede-Swagemakers 1997 UK Heart Attack Study 1998) An analysis of out-of-hospital mortality from AMI in Pennsylvania in 1998 showed that in relatively rural areas where travel time to the hospital was 25 minutes or longer 72 of all AMI deaths occurred outside the hospital compared with a statewide average of 49 of AMI deaths occurring outside of the hospital (OrsquoNeill 2003) In New York State 55 of IHD deaths occur outside the hospital compared to 35 of stroke deaths (NYS DOH 2009b)

Hudson River PCB Contamination

During an approximate 30-year period ending in 1977 the General Electric (GE) plants in Hudson Falls and Fort Edward used PCBs in the manufacture of electrical capacitors and discharged large quantities of PCBs into the upper Hudson River Estimates of the total direct discharges to the river are as high as 1330000 pounds In 1983 the US EPA classified a 200shymile section of the Hudson River from Hudson Falls to New York City as a National Priority List site making it one of the largest Superfund sites in the United States In 2002 the US EPA issued a Record of Decision calling for targeted environmental dredging and removal of PCB-contaminated sediment in the upper River (USEPA 2002) The dredging began in May 2009

The Superfund environmental investigations and planned remedial actions have focused on the areas of the river near Hudson Falls and Fort Edward where the highest levels of PCBs in sediments are located The 2002 Record of Decision provides information about the human health risk assessment conducted for the Upper and Mid-Hudson and the ecological risk assessment conducted for the Upper and Lower Hudson River (USEPA 2002) Fish contamination and fish consumption are the primary concerns at the site

In 1976 fishing was banned in the upper river and in 1995 a catch and release policy replaced the ban NYS DOH continues to recommend people eat no fish from the Upper Hudson River that children under age 15 and women of childbearing age eat no fish from the entire 200 mile stretch of river below Hudson Falls and that the general population eat none of most species of fish caught between the federal dam at Troy and Catskill approximately 40 miles south of Troy The Lower Hudson remains closed to commercial fishing for striped bass and eight other species

The fishing bans and advisories were issued because ingestion of fish is the major potential pathway for exposure for the general population Many previous studies have shown relationships between consumption of fish from contaminated waters and human PCB levels Another potential pathway for exposure less frequently studied to date is inhalation of PCBs NYS DOH and ATSDR launched ldquoThe Hudson River Communities Projectrdquo in 2000 to address this issue (More information on this project is provided below)

POPS and PCBs

The category ldquopersistent organic pollutantsrdquo (POPs) includes man-made chemicals that are by definition persistent in the environment Examples of POPs are polychlorinated biphenyls

5

(PCBs) polychlorinated dibenzo-p-dioxins (PCDDs) polychorinated dibenso-furans (PCDFs) and pprsquo-diphenyldichloroethene (DDE) a breakdown product of the pesticide dichlorodiphenyltrichloroethane (DDT) These chemicals are stored in human fat tissues accumulating over an individualrsquos lifetime They also accumulate in fish and other species thereby concentrating up the food chain Levels of most POPs measured in blood serum of Americans have decreased in the last few decades An exception is polybrominated diphenyl ethers (PBDEs) levels of which have been increasing in recent years While levels of many POPs are declining a large number continue to be detectable in blood for most of the US population (USDHHS 2009)

This follow-up investigation focuses on the Hudson River where the POPs of concern are PCBs PCBs are a group of synthetic chlorinated compounds manufactured in the US through 1977 Production was halted due to concerns about the persistence of PCBs in the environment and the toxicity of PCBs shown in animal models PCBs were used primarily by the electrical industry for capacitors and transformers and were also used in hydraulic fluids fluorescent light fixtures flame retardants inks adhesives carbonless copy paper paints pesticide extenders plasticizers and wire insulators PCBs can be oily liquids or solids and some are volatile and can exist as vapor in air PCBs in the environment are generally mixtures of the many different types known as congeners (ATSDR 2000)

The variety of types of PCBs adds to the complexity of studying adverse animal and human health effects Increasingly PCB research focuses on specific PCB congeners and groups of congeners because specific types of congeners appear to have different mechanisms of action and toxicity Dioxin-like PCBs (DLPCBs) are an important subset of PCBs due to concerns about their toxicity The focus on dioxin-like PCBs is paralleled by increasing research on the group of POPs with dioxin-like properties This type of PCB and POP research assigns toxic potency values relative to dioxin to various PCBs and POPs to provide an overall toxic equivalency value to specific PCB and other POPs

From 2002 through 2007 NYS DOH and SUNY Albany School of Public Health researchers conducted the Hudson River Communities Study among residents of Hudson Falls and Fort Edward the villages where the PCB discharges from two GE electrical capacitor plants occurred The comparison community was upriver from the discharges in Glens Falls This studyrsquos evaluation of exposures included measuring PCBs in outdoor air and in blood serum The study found that PCB levels in outdoor air were statistically significantly higher in the Hudson FallsFort Edward area (072 nanograms per cubic meter) than in the upriver Glens Falls area (040 nanograms per cubic meter) but that the PCB concentrations in the Hudson FallsFort Edward area were lower than those in other communities with known PCB-contaminated sites and similar to levels reported in other locations in the northeastern United States A statistically significant reduction in the total PCB concentrations in outdoor air samples was observed at a distance of 1200 meters (34 mile) from the river from 0760 ngm3 to 0497 ngm3 (Fitzgerald et al 2007 Palmer et al 2008)

The average PCB concentrations in serum for the study and comparison populations did not differ significantly (307 ppb wet weight versus 323 ppb) The average toxic equivalency quantities (TEQ) for dioxin-like PCB congeners for the study and comparison area populations

6

also did not differ In addition the study found no detectable differences in study subjectsrsquo serum PCB levels associated with proximity to contaminated portions of the river or other PCB sources The study did find that lifetime consumption of Hudson River fish was associated with higher serum PCB levels (Fitzgerald 2007 Palmer 2008) Another recent biomonitoring study of Brooklyn Manhattan and New Jersey fisherman who consumed versus those who did not consume their catch from the lower Hudson River and New YorkNew Jersey harbor suggests eating fish from these waters is not associated with an increased body burden of PCBs or most other organochlorines This study did not address the inhalation pathway (Morland 2008)

PCBs and Cardiovascular Disease Outcomes

Research studies on adverse health effects associated with PCB exposure have focused on many types of outcomes organs and biological systems including cancer liver reproductive and developmental effects thyroid endocrine immune system and neurological effects as well as specific effects on the skin and eyes Relatively little research has focused to date on associations between PCB exposures and CVD outcomes The majority of human studies have assessed health effects in occupational groups having relatively high-level exposures among people consuming fish from contaminated waters or people exposed to relatively high levels due to accidental contamination of food (USDHHS NIH 2010)

PCBs have been shown to produce a variety of toxic effects in animals These effects include cancer allergies hypersensitivity damage to the central and peripheral nervous systems reproductive disorders and disruption of the immune system (ATSDR 2000) In contrast research to date does not tell us whether specific types of human health effects have occurred from the low-level exposures experienced by the general population Some human epidemiological studies of relatively low levels of PCDDs PCDFs and PCBs suggest possible relationships not only with neurodevelopment of infants but also diabetes and a condition called insulin resistance conditions that increase CVD risk (Longnecker 2001 Carpenter 2006 Arisawa 2005)

In 1998 a study of a population in Seveso Italy exposed to relatively pure TCDD (dioxin) from an industrial accident showed increased mortality from CVD respiratory diseases and diabetes (Bertazzi 1998) These findings suggested the possibility that other dioxin-like chemicals such as dioxin-like PCBs might increase risk for these diseases Some more recent animal studies described below suggest possible mechanisms of action for PCB effects that increase risk for CVD diabetes and hypertension Another relevant source of information is from research using the biological monitoring data from the National Health and Nutrition Examination Survey (NHANES) project to look at the association between PCB levels in blood and self-reported diabetes and heart disease The body of evidence based on analytical human epidemiological studies remains relatively sparse however

Some recent studies using NHANES biomonitoring data show associations between PCB levels and CVD (Lee 2006 Lee 2007 Ha 2007 Everett 2008) These studies are cross-sectional (comparisons at one point in time) evaluations of PCB levels measured in blood samples from individuals and self-reported CVD and diabetes a risk factor for CVD While these studies use individual-level biomonitoring and self-reported health outcome data they are considered to be

7

relatively weak in terms of providing evidence of causality because both exposure and outcome are measured at the same point in time Given the nature of NHANES data collection and reporting practices these studies are not able to control for possible geographical confounding which may occur if urban versus rural residence is associated with higher PCB levels and higher risk for CVD for example The higher PCB levels among African Americans in these data suggest urban residence may be playing a role in the PCB data

The results of animal studies suggest multiple mechanisms by which PCB exposures could affect cardiovascular risk as well as multiple types of adverse health outcomes One hypothesized mechanisms of action specifically for dioxin-like PCBs is that they may have anti-estrogenic effects which impact CVD risk factors such as serum cholesterol levels and blood pressure (Lind 2004) Several studies suggest PCBs interact with a specific receptor (AhR) and increase cellular oxidative stress causing inflammation cell injury and cardiovascular dysfunction (Hennig 2002) Similarly some studies suggest PCB-induced inflammation of fat cells leads to obesity and increased risk for coronary artery disease (atherosclerosis) (Arsenescu 2008)

3 FOLLOW-UP STUDY DATA SOURCES AND METHODS

This follow-up investigation will provide perspective on the 2004-2005 findings (Sergeev and Carpenter 2005 Shcherbatykh et al 2005) by producing results from analyses conducted at a more detailed level of geography the block group rather than ZIP code in order to better control for confounding When the term confounding is used by epidemioogists it refers to the problem that can arise when there is a factor such as income level that is not the factor of interest in the study but that is related to the risk factor being studied (residential location ndash proxy for exposure) and to the disease being studied (CVD) This type of factor is labeled a ldquoconfounderrdquo because it can confound ie interfere with or confuse the evaluation of associations among the variables of interest in the data All types of epidemiological studies face issues of confounding but ecological ie group-level analyses face additional challenges related to potential confounding because analyses can not account for some important individual differences within the groups that are the units of analysis (Morganstern 1998b Hertz-Picciotto 1998) This investigation was able to control for each individualrsquos age sex and race but not for individual-level risk factors such as income or education level

The follow-up investigationrsquos specific objectives are to

improve the ability to control for confounding by socio-economic status through the use of block groups rather than ZIP codes as the unit of analysis

compare the hospitalization rates among residents of block groups nearest the Hudson River to those among residents of block groups farther from the Hudson River

compare the findings of this follow-up study to those of the earlier studies

Study Area and Exposure Areas

The overall study area is comprised of the 12 counties with geographic areas abutting the Hudson River south of the source of PCB contamination from the General Electric facility in Hudson Falls The 12 counties are Westchester Rockland Putnam Orange Dutchess Ulster

8

Columbia Greene Rensselaer Albany Washington and Saratoga US Census boundaries are used to define the study area because census population estimates provide the data for the study area populationrsquos size and the populationrsquos age sex and race distribution Because the study period spanned 1990 through 2005 and populations change over time 1990 and 2000 Census data were used to linearly interpolate the between-year populations from 1991 through 1999 for each block group 2000 data were used without extrapolation to estimate block group populations from 2001 through 2005

Because block group boundaries change between the Census years 1990 block populations were normalized to fit 2000 block group boundaries before the estimation of sex age and race group specific populations In the digital map 2000 block group boundaries were overlaid on the 1990 block centroids to assign 1990 block population counts to the corresponding 2000 block group boundaries Then 1990 block populations were summed to fit the block group boundaries Next population counts by age group (0-14 15-24 25-34 35-44 45-54 55-64 65-74 75+) sex (male female) and race (white black asian other) were estimated for each year at the block group level This investigation includes all ages over 25 years and the study population is divided into four race categories white black or African-American Asian and ldquoOtherrdquo The Pacific Island group was combined with the Asian group and American Indian ldquootherrdquo and ldquomultiple racesrdquo categories are combined into a larger ldquootherrdquo category

Classification of the population into potential exposure categories for this study is based exclusively on location relative to the Hudson River Two types of exposure proxy categories were developed block group adjacency and block group centroid distance to the river shore For the adjacency categories block groups were assigned to the adjacent category if they directly abutted the Hudson River and all other block groups in the 12 counties were assigned to the nonshyadjacent category

For the distance categories a central point (centroid) within each block group was identified and used to assign block group centroid distance to the river shore The block group centroids were population-weighted which means the geographic distribution of households within 2000 census blocks were used to locate block group centroids based on the geographic population distribution By locating a population-weighted block centroid rather than a more straightforward distance centroid the assignment of block group distance to the river shore more accurately reflects distance to the river for the majority of block group residents Buffers of one-half mile and one mile from the Hudson River boundary were created for the assignment of each block group into one of the three distance categories less than frac12 mile (closest) frac12 to one mile (close) and farther than one mile from the river (distant)

Distance categories were selected a priori at the start of the study Because environmental sampling data were not collected along the length of the river included in this investigation we are not able to determine whether these distance categories represent actual differences in exposures As discussed previously a statistically significant reduction in the total PCB concentrations in outdoor air samples was observed at a distance of 1200 meters (34 mile) from the river from 0760 ngm3 to 0497 ngm3 in a study conducted in the Hudson Falls Fort Edward area near the source of PCBs (Fitzgerald 2007 Palmer 2007)

9

For the adjacency categories of the 2371 block groups in the study area 255 (11) were classified as adjacent to the Hudson River and 2116 (89) were classified as non-adjacent (Figure 2) For the distance categories 225 (9) were classified as closest to the river 216 (9) as close and 1930 block groups (81) as distant from the river

Comparison of Figures 1 and 2 show that block groups include smaller geographic areas than ZIP codes and that block groups are relatively smaller in area in urban versus rural areas The study area includes a total of 368 ZIP codes compared to a total of 2371 block groups ZIP code average population in the study area is approximately 8000 compared to block group average population of 1200 Comparison of Figures 2 and 3 shows that using population-weighted centroid distance to the river produces different exposure areas than adjacency Fifty-three (20) of the adjacent block groups were categorized for the distance indicator as distant because the population in the block group was concentrated in the area farther than one mile from the river On the other hand about 11 of the 2116 non-adjacent block groups were classified as closest (77) or close (162)

Hospitalization Data

Hospitalizations for ischemic heart disease (IHD) and cerebrovascular disease (stroke) were included in this study More specifically hospital discharge records that listed as the principal diagnosis either IHD (ICD-9-CM 410-414 and 4297) or cerebrovascular disease (ICD-9-CM 430-436) and subsets of each of these categories similar to the subsets evaluated in the Sergeev et al and Shcherbatykh studies were included The subsets are AMI and angina subsets of IHD and ischemic and hemorrhagic stroke which are subsets of cerebrovascular disease (stroke) (Principal diagnosis represents the illness for which the person was admitted to the hospital) Angina is an outcome not evaluated in the prior studies It was added here for comparative purposes because it represents an outcome appropriately addressed by primary outpatient rather than inpatient hospital care Angina is one of a group of outcomes included in the recently developed national ldquoPrevention Quality Indicatorsrdquo (USDHHS AHRQ 2007) as an indicator for geographic areas or populations for whom primary care is inadequate Table 3 lists the specific ICD-9-CM codes and definitions included in this investigation

While one of the previously published studies (Huang et al 2006) showed elevations in hypertension (high blood pressure) rates this outcome is rarely listed as the principal reason for hospitalization As a secondary diagnosis however hypertension is extremely common with substantial overlap with the primary diagnoses of IHD and stroke In addition hypertension hospitalizations were not as elevated in the ZIP codes adjacent to the Hudson River as the IHD and stroke outcomes As a practical matter geocoding the hypertension hospitalization data (1364000 hospitalizations) would overwhelm the resources for this project For these reasons hypertension was not included in this follow-up investigation

The source of the hospitalization data was the NYS DOH Statewide Planning and Research Cooperative System (SPARCS) established in 1979 to collect detailed records on discharges from hospitals located in New York State Persons seen in the emergency department but not admitted are not included in the SPARCS data Hospitalizations at US Veterans Administration

10

Hospitals are also not included in this dataset We identified the hospital transfers in the data and excluded them so that individuals are only counted once for the same event

The NYS DOH Data Protection Review Board for SPARCS hospitalization data approved our use of hospitalization records for this investigation and provided approximately 460000 records with a principal diagnosis of either IHD or stroke Because of the large number of cardiovascular hospitalizations and this studyrsquos requirement for geocoding each hospitalized personrsquos address we selected a 50 random sample of the hospital discharges with principal diagnoses of IHD or stroke for inclusion in the study (229948 records) Approximately one percent of these records (2737) were eliminated from the study because they contained no residential address resulting in 227211 hospitalization records 24939 of these records were eliminated after geocoding because they were not within the 12 county study area This left 202272 records remaining in the study

For geocoding to the block group level the hospitalization street addresses were modified as needed using US Postal Service standards The addresses were assigned geographic coordinates using commercially available geocoding software (MapMarker 2004) A combination of land parcel data and street centerline files were used to assign geographic coordinates No contact was made with cases parents legal guardians or next of kin of cases to determine residential locations

Once geographic coordinates were assigned to cases through address-matching the case locations were overlaid onto digital maps of the study area using a geographic information system so that the number of observed cases falling within the study area boundaries could be determined and a 2000 Census block group could be assigned to each hospitalization In order to protect confidentiality no maps of individual case locations are provided Geographic coordinates were not able to be assigned for some addresses for example rural routes or post office boxes Internet searches were conducted for addresses that were located in facilities such as nursing homes and trailer parks

While fewer than one percent of records were missing sex or age information approximately 15000 records seven percent were missing race information Evaluation of the geographic and temporal distribution of these records showed they were not randomly geographically distributed 47 of the records with no race information were from one hospital for that hospital race was missing from 100 of the records in all years except one after 1997 mid-way through the study time-frame Evaluation of other records missing race showed similar patterns of 100 of records missing race in specific years In order to prevent biased hospitalization rate estimates records missing race were assigned race randomly assuming a race distribution equal to that of the individualrsquos residential block group

Statistical Analysis

This investigation used negative binomial multivariable regression models the statistical methods used in the previous studies (Sergeev and Carpenter 2005 [IHD] Shcherbatykh et al 2005 [stroke]) The follow-up studyrsquos analyses compared hospitalization rates for populations in block groups adjacent or in close proximity to the Hudson River to rates for populations more

11

distant from the Hudson River while adjusting for the potential influence of other factors Sex age and race-specific hospitalization count data comprise the dependent variable and the statistical methods attempt to predict the counts using the sex age and race distributions from Census data as well as block group-level demographics such as median income Models were first estimated using basic Census information for age sex and race Then more fully adjusted models were estimated using information at the block group level for income education population density Hispanic ethnicity and distance to the nearest hospital

More specifically each hospitalized personrsquos block group was used to assign quartile categories of block group median household income of population with less than a college education population density of the population identified as Hispanic and distance from the block group centroid to the nearest hospital These variables were selected in advance The population density variable was selected to account for potential effects on hospitalization rates of large differences in population density from the relatively rural northern counties to the more urban southern counties in the study area of the population identified as Hispanic was included because Hispanic ethnicity may be associated with CVD risk but Hispanic ethnicity was not available in the hospitalization records Distance to the hospital was included because studies show that shorter distances increase the likelihood that individuals with AMI and stroke survive to be admitted to the hospital

As described previously the literature about risk factors for CVD shows strong associations between lower income and education and higher CVD risk Regarding the effects of income and education on hospitalization rates the literature from analytic studies specifically of hospitalizations for CVD is quite sparse Most analytic studies focus on prevalence incidence or mortality in population-based studies or cohort studies and there are not many of these An example of a study that was useful as background is a study of income level and CVD mortality that showed strong associations of lower income and higher CVD mortality for men and women with the relationship for men being linear throughout the range of income levels while for women the relationship was only strong at levels of income below the median (Rehkopf 2008)

This follow-up studyrsquos approach was to use the same set of variables for the more fully adjusted models using quartile cut points for all the sexrace groups and types of CVD diagnostic subsets The 2000 Census data were used to create the population-weighted quartile categories to be used in the statistical analyses The quartile cut points are shown below

population density(persons per square mile)

median 1999 household income

with less than a college education

with Hispanic ethnicity

Hospital distance

Quartile 1 61 - lt507 $2499-lt$40968 0 - lt31 0 - lt2 lt25 (median) Quartile 2 507 - lt102 $40968 - lt54650 31 - lt43 2 - lt5 25 - lt6 (50-75) Quartile 3 102 - lt6410 $54650 - lt72944 43 - lt55 5 - lt10 6 ndash 11 (75-90) Quartile 4 6410 - 94372 $72944 - 200001 55 - 100 10 - 86 11 ndash 45 (90-100)

SAS version 9 provided the statistical programs used to estimate hospitalization rates for exposure proxy categories and demographic categories The models provide ldquocontrast

12

estimatesrdquo ie rate ratios for one category versus a reference category of a variable and 95 confidence intervals (CI) for these rate ratios For the location categories rate ratios for the areas adjacent are compared to areas not adjacent to the river (reference) and areas closest and close to the river are compared to areas distant from the river (reference) If a rate ratio is greater than 100 the model has estimated an elevation of hospitalizations in that category of the study population compared to the reference population If the rate ratio is less than 100 then there are fewer than expected hospitalizations compared to the reference population

The magnitude of the elevation or deficit is also estimated by the rate ratio For instance if the hospitalization rate is twice as high in the adjacent population it would result in a rate ratio of 200 while a 50 higher rate would result in a rate ratio of 150 If the hospitalization rate in the comparison group was only half as big as in the reference group this would result in a rate ratio of 050 The 95 confidence interval assists with determining whether the observed differences are statistically meaningful ie statistically significant The confidence interval provides the range in which there is a 95 probability of including the true rate ratio

In addition to directly assess potential confounding we examined the associations between income and exposure proxy categories and between income and CVD hospitalization risk Because the modeling process provides rate ratios that provide comparisons among groups but not the rates themselves we also calculated the rates per population for each sex race and income category in order to directly observe the relative magnitudes of CVD hospitalization rates Finally analyses stratified by income quartile were also conducted in order to assess the potential role of confounding by income The stratified analyses consist of negative binomial regressions conducted separately for each quartile of income

4 RESULTS

Demographics Associations with Proximity to the River

Study Population Table 4 provides detailed demographic characteristics for the total study population and the populations of the adjacent versus non-adjacent and nearer versus farther from the river block groups from the 2000 Census The total population of the study area for this one year of the study time period is 2924631 persons The first row of Table 4 shows that approximately 11 percent of the study area population is classified as living in block groups adjacent to the Hudson River and 89 live in non-adjacent block groups Approximately 8 of the studyrsquos population lives in the closest block groups approximately 10 lives in close block groups and approximately 82 in distant block groups The table also shows that approximately 81 of the total study population is identified as white 9 black 3 Asian less than 01 Pacific Islander 02 American Indian 4 ldquootherrdquo and 2 ldquomultiple racesrdquo

The 2000 Census data (Table 4) also show that the populations in block groups adjacent and nonshyadjacent to the river are similar in terms of the percent of minority population but the adjacent block groups include a smaller percentage of persons who have attended college and a higher percentage of households below the poverty level Median income is $47320 for block groups adjacent versus $62668 for block groups not adjacent to the river Data for the study area from the 1990 Census show similar distributions (data not shown)

13

For the distance to the river categories there is a gradient for the three levels of distance from the river with the closest block groups on average having higher percentages of minorities smaller percentages of persons who have attended college and higher percentages of households below the poverty level (Table 4) The populations of the close and closest block groups are more similar to each other than to the populations of the distant block groups The population of the closest block groups is estimated to be 19 black close block groups 14 black and distant block groups 8 black The proportions for the ldquootherrdquo category used in the analyses which includes multiple races and American Indians are approximately 12 in both the closest block groups and in the close block groups compared to 5 in the block groups most distant from the river Median income for the closest block groups is $40748 close block groups $44962 and most distant block groups $64847

The bar chart below illustrates the correlations between US Census 2000 indicators of block group socio-economic status and block group centroid (population weighted) distance from the river for the study population

2000 Census Demographics for Study Population by Distance to River Category

black other hispanic no college below poverty

of pop ulati on

0

10

20

30

40

50

60

lt12 mile

12 to 1 mile

gt1 mile

Table 5 shows selected demographic information for the person-year estimates used in the study analyses for the 1990-2005 study period Person-year data represent individuals from Census estimates counted repeatedly for each year of the study timeframe As described in the Methods section the estimation of person years used interpolation to account for population changes from 1990-2000 and techniques for taking account of block boundary changes The person-years estimates for demographic categories (complete data not shown) show similar patterns as the 1990 Census data (data not shown) and 2000 Census data (Table 5) Table 5 for example shows nearly identical percentages of population in each of the river distance categories as Table 4

Table 5 also shows person year information for income quartiles and exposure proxy categories as well as race and exposure proxy categories that is consistent with the Table 4 data for the

14

2000 population Table 5 row percentages show for example that while approximately 11 of the total study population live in adjacent block groups approximately 19 of the population in the lowest income quartile live in adjacent block groups By design approximately 25 of the total study population is in each of the income quartiles Table 5 column percentages show however that in the adjacent block groups 41 of the population is in the lowest income quartile and only 9 are in the highest income quartile

For the block groups within frac12 mile of the river the distribution among income quartiles is even more skewed towards the lowest income quartile 61 of the population living within frac12 mile of the river is in the lowest income quartile and only 6 is in the highest income quartile Given that approximately 25 of the total study population is within each income quartile these percentages indicate that living within the block groups closest to the river is associated with more than a doubling of the likelihood of being in a block group in the lowest socio-economic status quartile 436 of those living in the block groups close to the river (12 to one mile from the river) are in block groups in the lowest income quartile While less than a doubling this indicates that living in the close block groups is associated with a 75 increase in the likelihood of being in the lowest quartile for socio-economic status

Hospitalizations Table 6 shows the process of removing hospitalization records from the study for a variety of reasons The first column includes the 202272 records that contained a residential address and were located within the 12-county study area as described in the Methods section The next five columns show the records remaining after data cleaning exclusion of transfers elimination of records with missing covariates and exclusions due to poor geocoding results Approximately 33000 records were eliminated because they were transfers from one hospital to another for the same event For approximately 1100 records geocoding was inadequate for assigning block group for imputing race as described in the Methods section 336 records were missing sex (11) or age (325) Of the 167733 records remaining 154220 (92) were able to be adequately geocoded to a street address for assignment in a specific block group While approximately eight percent of hospitalizations were eliminated due to inadequate geocoding Table 6 shows that the percentages of hospitalization records for white black Asian and ldquootherrdquo categories and the distribution among diagnostic categories change very little after these records were dropped

The final column of Table 6 shows that of the 154220 hospitalizations in the analysis 101012 IHD hospitalizations comprise the largest health outcome category and are 65 of the hospitalizations in the study AMI hospitalizations (35811) account for 35 of the IHD hospitalizations and angina (4731) accounts for five percent of the IHD hospitalizations Stroke hospitalizations (53208) comprise 34 of the studyrsquos hospitalizations Approximately 36 of the stroke hospitalizations are for ischemic stroke (19371) and approximately 11 are for hemorrhagic stroke (5995)

Table 7 shows the distributions of hospitalizations among the four income quartiles While each income quartile (by definition) contains approximately 25 of the study population for all types of hospitalizations we see that the lowest income quartile includes more than 25 of hospitalizations from a minimum of 286 (IHD) to a maximum of 356 of hospitalizations (angina) The highest income quartile includes fewer than 25 of hospitalizations for each CVD

15

category ranging from 182 for angina to 224 for hemorrhagic stroke These deviations from the 25 quartile distribution show a strong linear pattern with the largest percentage of hospitalizations coming from residents of block groups in the lowest income quartile and the smallest percentages coming from residents of the most wealthy block groups For AMI the lowest income quartilersquos share of hospitalizations is 50 higher than the highest income quartilersquos share of hospitalizations (30 vs 20) IHD and stroke show similar differences and angina shows a doubling with the lowest income quartile contributing 36 of hospitalizations compared to the highest income quartilersquos 18 share of hospitalizations

Table 7 also shows the distribution of hospitalizations among the race categories and the study population race category distribution for comparison Blacks show lower percentages of IHD and AMI than their population percentages and higher percentages of angina and stroke While full presentation of race and sex findings is beyond the scope of this report the lower than expected percentages of IHD and AMI among blacks who on average have lower socioshyeconomic status and are at greater risk for CVD hospitalizations raised questions that were addressed by calculating hospitalization rates per 1000 population for each sex and race group for each income quartile (Appendix B Table 2) Figures 4 5 and 6 show these rates for the white black and Asian categories (These rates for the ldquootherrdquo category are comparatively very high and are not shown on these graphs This issue is addressed in the Discussion section) Figures 4 5 and 6 show the consistent and strong effects of income quartile on hospitalization rates for AMI and stroke for whites blacks and Asians of both sexes The other outcomes show similar patterns (The ldquootherrdquo group showed an opposite pattern [See Discussion Section])

Demographics Summary

The strong associations of lower income with residence closer to the river (Table 5) and lower income with higher hospitalization rates (Tables 7 amp 8) create a difficult challenge for this investigation The multivariable analyses of hospitalization rates attempt to control for socioshyeconomic differences However because these income differences which strongly affect hospitalization rates are also strongly correlated with the gradations of exposure (river distance) it will be difficult to disentangle the influence of socio-economic status from the influence of potential exposures

Statistical Analyses Proximity to the River and CVD Hospitalization Risk

All the variables described above were included in the multivariable regression models and almost every quartile category for every variable showed statistically significant and expected types of effects for all the CVD hospitalization categories (Appendix B Tables 1a and 1b) The following presentation of results focuses on the rate ratios for the exposure proxy variables Table 8a (IHD) and 8b (stroke) show the regression modeling results Results are shown first for models that include only age sex and race in addition to the exposure proxy categories and second for models that also include median income education population density Hispanic ethnicity and distance to nearest hospital The tables provide numbers of hospitalizations and person-years to assist with interpretation For example for total IHD we can see that while 108 of the study population lives in adjacent block groups 122 of the hospitalizations are among people in adjacent block groups This difference of 14 percentage points is the basis for

16

the finding of a statistically significantly elevated rate ratio (risk) for IHD among people living in the adjacent block groups (RR=133 CI 112-157)

Looking across the columns of Tables 8a and 8b we can see the rate ratios for each of the outcomes for the adjacent versus non-adjacent block groups and the results are similar to those for IHD The elevated rate ratios range from 123 for angina to 136 for AMI The only rate ratio not elevated in these first models adjusting only for age sex and race is for hemorrhagic stroke Going down the rows of the Table the next set of rate ratios are for the distance to the river categories The findings are very similar to those for the adjacent block groups Here however we have two exposure proxy categories allowing comparison of the two These rate ratios show no evidence that living in the block groups closest to the river (lt12 mile to the river) produces greater risk than living in the block groups close to the river (12 to 1 mile from the river)

The next set of rate ratios are for models with additional adjustment for the demographics of each hospitalized personrsquos block group as well as distance to the nearest hospital As expected adjustment for block group median income and other demographics accounts for some of the variability in hospitalization counts and thereby reduces the magnitude of the rate ratios estimated for the river adjacency and distance categories The rate ratio elevation for total IHD for example is reduced from 133 to 107 (RR=107 CI 104-111) The elevated rate ratios in these more fully adjusted models for the adjacency analyses are of similarly small magnitude ranging from 106 for AMI to 110 for ischemic stroke Four of the six outcomes show statistically significant elevations Again only hemorrhagic stroke shows no evidence of an elevated rate ratio

For the river distance categories none of the rate ratios show any evidence that living in the block groups closest to the river (lt12 mile to the river) produces greater risk than living in the block groups close to the river (12 to 1 mile) Of the six types of outcomes only two show statistically significantly elevated rate ratios for living in the closest block groups (total stroke and ischemic stroke) For the close block groups four out of six outcomes show statistically significantly elevated rate ratios The statistically significantly elevated rate ratios vary from 106 (AMI stroke) to 113 (angina)

To address the strong evidence described earlier for confounding in these data ie the association of lower income with both CVD hospitalizations and residence closer to the river we also conducted the analyses separately for each quartile of income These results are presented in Table 9 The first results are from analyses adjusting only for age sex and race and the second set of results are from models including the additional demographic factors For simplicity rate ratios are shown only for block groups closest to the river compared as in the other models to block groups farther than one mile from the river (Results were similar for the other distance categories (data not shown)

For the two lowest income quartiles Table 9 shows results similar to those for the study population as a whole (Table 8a 8b) However for the two highest income quartiles which represent frac12 of the study population Table 9 shows a consistent pattern of reduced rate ratios for residence in the closest block groups within frac12 mile of the river In the highest income quartile

17

group centroid (population-weighted) distance to the river shore was used to assign hospitalizations in each block group to one of three categories less than one-half mile from the Hudson River (closest) between one-half to one mile (close) from the river and farther than one mile from the river (distant) Multivariable regression analyses were conducted to compare rates of hospitalizations among the exposure categories assigned at the block group level These analyses controlled for age sex and race using individual-level data and population density percent Hispanic median income education and distance to the nearest hospital at the block group level

Results The use of block groups resulted in 2371 block groups as the units of analysis for the study rather than the 368 ZIP codes in the 12-County study area The average population per block group in the study area is 1200 compared to average population of 8000 per ZIP code The smaller geographical area and smaller populations in block groups contributed to improving the accuracy of the assignment of income and other group-level demographics to the hospitalizations included in the study and enhancing control for confounding by socio-economic status (Tables 8a and 8b)

The studyrsquos multivariable analyses showed relatively small but statistically significant elevations of CVD hospitalization risk for residents of block groups categorized as adjacent in close and closest proximity to the Hudson River There is no evidence that residents of the closest block groups (less than frac12 mile) had higher hospitalization rates than residents of the close block groups (between frac12 to 1 mile) If there were a strong association between exposures from living near the river and the risk of CVD hospitalization such a ldquodose-responserdquo might have been shown by the analyses

The data showed strong statistically significant associations between CVD risk and socioshyeconomic status To enhance the control for socio-economic status separate analyses for each income quartile were conducted These stratified analyses showed that residence in block groups near (close closest or adjacent to) the Hudson River was associated with increased risk for CVD hospitalizations among residents of lower income block groups Conversely residence in block groups near the Hudson River was associated with decreased risk for CVD hospitalizations among residents of higher income block groups In other words in analyses that held income constant by looking only within the lower or higher income populations living near the river shows opposite effects depending upon which income group is considered The estimated increased and decreased risks are of sufficient magnitude to cancel each other out when average rate ratios for the four income quartiles are calculated The average rate ratios can be interpreted as showing no evidence that hospitalizations among the general population are elevated for residents living near the river

This type of group-level statistical analysis is not able to precisely identify the reason for these differing results However the strength and direction of associations among variables in the statistical models can be evaluated in light of known risk factors to suggest likely explanations In this case socio-economic status appears to be the most likely explanation for the findings The strong association between socio-economic status and residential location (the indicator of potential exposure) and between socio-economic status and CVD hospitalization risk makes it likely that the small but statistically significant elevations shown in the multivariate models

iii

(Tables 8a amp 8b) are the result of the inability to completely control for socio-economic statusrsquo effects on CVD hospitalization risk The strong effect of socio-economic status on residential location (exposure proxy) and on CVD hospitalization risk makes it likely that these overall multivariable regression results are affected by uncontrolled ie residual confounding by socioshyeconomic status

Because of the similarities in study design and statistical analysis the prior studyrsquos findings can be compared to some degree with the findings of the multivariate regression analyses in the current study The major difference between the studies is the composition of the comparison areas for the analyses Because of this difference as well as other differences described in the report the studies are not entirely comparable

CVD hospitalization rate ratios estimated for living in block groups adjacent closest or close to the river in the current study are substantially lower than the rate ratios for residence in ZIP codes adjacent to the Hudson River in the previously published studies In most cases the current analyses show elevations below 10 in contrast to elevations of 20 to 39 in the prior studies The results of the current study are quite similar to the prior study results in terms of the associations reported for age sex race and income These similarities and the general methodological similarities between the current and prior studies suggest the likelihood that the prior studiesrsquo findings also were affected by confounding by socio-economic status

Conclusions and Recommendations While this type of study can neither prove nor disprove causal links among risk factors and health effects it is useful for examining evidence for unusual patterns that might warrant additional investigation This evaluation of hospitalization data residential location and socio-economic factors in block groups in the 12 Counties abutting the Hudson River revealed a closely coinciding pattern of lower socio-economic status and residential proximity to the Hudson River the proxy for potential exposure and a strong association of lower socio-economic status and risk for CVD hospitalizations The strong associations among residential location socio-economic status and CVD risk prevent us from drawing definitive conclusions from the multivariable regression analyses that showed modest but statistically significant elevations of CVD hospitalization risk for residents of block groups near the river

To draw more definitive conclusions particularly for outcomes such as CVD that are determined by multiple known risk factors studies seeking to investigate the role of PCB or POP exposures will require resource-intensive methods that include gathering individual-level exposure or biomonitoring information as well as medical histories and information on other risk factors

This current investigation did not include environmental or biological sampling data One study did gather such data in the Hudson River region with the highest potential exposures (Hudson Falls Fort Edward) This study detected a statistically significant reduction in the total PCB concentrations in outdoor air samples at a distance of 1200 meters (34 mile) from the river However the PCB levels in outdoor air in this area were lower than those in other communities with known PCB-contaminated sites and similar to levels reported in other locations in the northeastern United States The study found no detectable differences in study subjectsrsquo serum

iv

PCB levels associated with proximity to contaminated portions of the river or other PCB sources (Fitzgerald 2007 Palmer 2008)

It is recommended that the results of the current investigation of CVD hospitalization rates and residence near the Hudson River be shared with NYS residents and other stakeholders with interest in this issue and be submitted for publication in a peer-reviewed journal

v

1 INTRODUCTION

The New York State Department of Health (NYS DOH) has a cooperative agreement with the Agency for Toxic Substances and Disease Registry (ATSDR) to perform environmental exposure and health assessments conduct health statistics reviews and perform epidemiological studies of populations in New York State which may have been exposed to environmental contaminants In October 2006 Member of Congress Maurice Hinchey requested that ATSDR conduct a health consultation to assess cardiovascular disease among people residing near the Hudson River A 200-mile segment of the Hudson River from Hudson Falls south to New York City is a National Priority List (NPL) (Superfund) site due to contamination of the river with polychlorinated biphenyls (PCBs) from GE sites in Hudson Falls and Ft Edward (See Appendix A for Member of Congress Hincheyrsquos letter)

In support of his request Member of Congress Hinchey cited three studies (Sergeev and Carpenter 2005 [ischemic heart disease] Shcherbatykh et al 2005 [stroke] Huang et al 2006 [hypertension]) These statewide studies suggest residents of ZIP codes containing hazardous waste sites with persistent organic pollutants (POPs) such as PCBs have elevated risk for hospitalizations for cardiovascular disease including ischemic heart disease stroke and hypertension Analyses conducted separately for ZIP codes next to the Hudson River suggest residents of these specific ZIP codes also have increased rates of hospitalization for cardiovascular disease

These studies published in 2005-2006 are ecological studies In an ecological study data are analyzed by evaluating risk factors and health outcomes among groups For example rates of disease are compared for populations of ZIP codes with varying median income levels to see if income levels appear to be related to disease outcome rates In these previously conducted studies rates of hospitalization for residents of ZIP codes with hazardous waste sites containing POPs were compared with hospitalization rates for residents of ZIP codes without such waste sites to look for evidence that potential exposures might be impacting health outcomes The studies report statistically significant elevations of cardiovascular disease hospitalizations in ZIP codes with POPs For coronary heart disease and a subset acute myocardial infarction the study showed elevations of approximately 15 and 20 respectively (Sergeev and Carpenter 2005) for stroke a 15 elevation (Shcherbatykh et al 2005) and for hypertension a 19 elevation (Huang et al 2006) in ZIP codes containing waste sites with POPs compared with hospitalization rates in other New York State ZIP codes (excluding NYC) that did not contain waste sites with POPs (Table 1)

In sub-analyses in these studies cardiovascular hospitalization rates for ZIP codes adjacent to the Hudson River were compared with cardiovascular hospitalization rates for ZIP codes in the rest of the state (excluding NYC) identified as not having inactive hazardous waste sites with POPs These sub-analyses showed statistically significantly elevated hospitalization rates of 36 and 39 for coronary heart disease and the subset acute myocardial infarction a 20 elevation for stroke and a 14 elevation for hypertension (Table 1) The Sergeev and Carpenter Shcherbatykh et al and Huang et al articles state that the Hudson River ZIP codes are in an area where people have higher income smoke less exercise more and have healthier diets This information is cited in these articles as evidence that known risk factors related to lower socioshy

1

economic status are likely not the cause for the elevated hospitalization rates in these Hudson River area ZIP codes

In these previously published reports Behavioral Risk Factor Surveillance System (BRFSS) data for entire counties are provided to support statements about the population of Hudson River ZIP codes having higher income and healthier lifestyles However because the ZIP codes adjacent to the Hudson River include only a proportion of the population of the counties abutting the river these county-level BRFSS data may be misleading regarding characteristics of the ZIP codes adjacent to the river (Figure 1) In fact data from the US Census show that the ZIP codes adjacent to the river have lower median income than the counties within which they are located The median income for the total population of the twelve counties adjacent to the Hudson River was $61010 in 2000 while the median income for the ZIP codes adjacent to the river was $45356 This median income for the population of the ZIP codes adjacent to the river is also slightly lower than the statewide (excluding NYC) median income of $47735 (US Census 2000) Thus the Hudson River ZIP codes analyzed in the Sergeev and Carpenter Shcherbatykh et al and Huang et al studies include a population of lower not higher socio-economic status compared to the rest of NYS (excluding NYC)

Because CVD and many CVD risk factors are associated with lower socio-economic status it is especially important to adjust as much as possible for socio-economic status when conducting comparative analyses of CVD hospitalizations or disease incidence or prevalence Regarding the hypothesis about POPs and CVD research studies described in more detail below have suggested possible biological mechanisms for POP exposuresrsquo effects on the heart There is very little research on human exposures to POPs and potential cardiovascular effects however

In response to Congressman Hincheyrsquos request for follow-up of the 2004-2005 findings ATSDR and NYS DOH agreed to conduct a similar study also ecological but at a different level of geography the Census block group By using a level of geography that is generally smaller in area and population than a ZIP code this follow-up seeks to more accurately adjust for socioshyeconomic and other factors that affect cardiovascular hospitalizations and to improve upon the geographic classification of relative distance to the Hudson River In addition rather than comparing hospitalizations among residents of Hudson River area ZIP codes to hospitalizations among residents of all other NYS (exclusive of NYC) ZIP codes with no waste sites with POPs this follow-up investigationrsquos analyses are restricted to the twelve counties abutting the Hudson river from Washington County south to Westchester County Within the study arearsquos twelve counties cardiovascular hospitalization rates in areas closest to the Hudson River are compared with rates in areas farther from the river This more localized approach may also assist with reducing the potential impact of regional differences in medical care practice or reporting factors that could affect the hospitalization data (Ko et al 2007) (Table 2 shows a more detailed listing of the methods used in the prior studies and follow-up study)

Regarding environmental exposures it is important to note that living near a hazardous waste site that contains PCBs or other POPs does not necessarily mean that human exposure from the site has occurred Depending upon the conditions at a site or the specific chemicals exposures may be highly improbable Information on body burden for example PCB levels in blood or on levels of the chemical in off-site air dust or soil is needed to support a claim of exposure The

2

studies published in 2005-2006 did not provide specific information such as environmental or biomonitoring data in support of the assumption that residence in ZIP codes containing hazardous waste sites with POPs or residence in ZIP codes adjacent to the Hudson River resulted in increased exposures to POPs in the population For the purpose of this follow-up we did not conduct environmental or human biological sampling to address this issue Rather this follow-up replicates the prior studiesrsquo approach with some enhancements to better control for confounding particularly confounding by socio-economic status

This descriptive epidemiologic investigation uses already existing hospitalization data from the Statewide Planning and Research Cooperative System and population data from the United States Census to evaluate comparative rates of cardiovascular hospitalizations While this type of investigation can not show proof of cause and effect it can assist in interpreting relationships among indicators of CVD burden and hypothesized exposures associated with distance from the Hudson River This descriptive investigation may also assist in determining whether additional health investigations of these associations are warranted

2 BACKGROUND

Cardiovascular Disease

The widely known risk factors for cardiovascular disease (CVD) in addition to age sex race and family history include tobacco use high blood pressure high serum cholesterol alcohol use obesity low fruit and vegetable intake physical inactivity diabetes low socio-economic status mental ill-health psychosocial stress and the use of certain medications (WHO 2010 NYS DOH 2009 Wing 1988 NYS DOH 2009a)

CVD Definitions CVD is a general term for any disease of the circulatory system The two main types of CVD are ischemic heart disease also termed coronary heart disease and cerebrovascular disease also called stroke Ischemic heart disease (IHD) results from restricted blood flow through the arteries supplying the heart muscle The most serious risk from IHD is acute myocardial infarction (AMI) another term for heart attack AMI occurs when blood flow to the heart is suddenly reduced or stopped usually due to a blockage in a coronary artery Angina is severe constricting pain in the chest often radiating from the area immediately over the heart to the left shoulder and down the arm Cerebrovascular disease refers primarily to stroke the sudden impairment of brain function resulting from interruption of circulation to the brain Ischemic stroke results from blockage of an artery supplying blood to the brain Hemorrhagic stroke occurs due to the escape of blood from a ruptured artery supplying blood to the brain

CVD Burden and Trends CVD is the leading cause of death for men and women of all ages in the United States and in New York State In 1999 for every death from IHD there were five hospitalizations for ischemic disease in NYS (NYS DOH 2009a) Mortality rates from CVD have been declining steadily since the 1980rsquos Research studies estimate that approximately half of the decline is due to improved treatments for CVD and its risk factors such as high cholesterol and high blood pressure and approximately half is due to decreasing incidence of CVD The decrease in incidence is due to trends in health behaviors primarily reduced smoking rates

3

CVD mortality rates do not appear to be declining for all segments of the population however and may actually be increasing among some groups particularly people of lower socio-economic status (Barnett et al 1999) This may be due in part to higher rates of smoking among people of lower socio-economic status as well as to other factors related to socio-economic status and race A growing literature points to multiple causes for CVD and other health disparities including health practices psychosocial stress limited resources discrimination and access to health care (Health United States 2007) Increases in the prevalence of diabetes and obesity in the general population partially offset advances in treatment and favorable trends for other risk factors An increasing ischemic heart disease mortality rate in women ages 35 to 54 since 1997 is attributed to these unfavorable risk factor trends (Ford 2007 Lloyd-Jones 2009) New York State (NYS) and other northeast states have high IHD and low stroke mortality with NYS having the highest IHD mortality and lowest stroke mortality among the 50 states (Howard et al 2009)

CVD Data Sources CVD mortality rates are more easily measured than incidence prevalence or hospitalizations because mortality data from vital records are a comprehensive and relatively accessible data source Information for estimating the incidence and prevalence of CVD outcomes comes primarily from health interview physical examination and record surveys Hospitalization data are available for NYS and the US but are generally collected and analyzed in order to track trends in hospital care and resources expended Actual CVD incidence data validated by physician diagnosis physician examination or reviews of medical records are not directly available for the entire US or for specific geographic areas of NYS Such information is available from population-based cohort studies that follow people over time in selected communities While the concern underlying this research project is to understand CVD levels in the community ie incidence and prevalence we do not have data that directly provide us with this information In the absence of such data hospitalizations are frequently used as indicators of the burden of disease in communities particularly in surveillance or hypothesis-generating investigations such as this one In NYS the Behavioral Risk Factor Surveillance System (BRFSS) an ongoing monthly telephone survey of adults aged 18 years and older provides information about the burden of CVD by age sex race income and education

CVD Risk Factor Data The BRFSS data for NYS show that over 20 of the population aged 65 and older reported having been diagnosed with CVD versus less than five percent of the population under 65 Males were slightly more likely to report having CVD than females Nonshyhispanic whites compared to other ethnic groups combined were slightly more likely to report having CVD None of these slight differences were statistically significant however The only statistically significant demographic risk factors in the BRFSS data were low income and education levels Statistically significant elevations of reported CVD were observed for respondents with incomes below $25000 and respondents with a high school degree or less education (NYS DOH 2009a)

Another factor that can affect the outcome measure hospitalizations is the likelihood of mortality prior to reaching the hospital Studies show a variety of factors affect whether an individual experiencing a heart attack or stroke outside the hospital survives to be counted in hospitalization data These factors include lack of knowledge about personal risk or prior disease socio-economic status race marital status and distance to the hospital (Galea 2007

4

Barnett 2006 Ayala 2003) Population-based studies report sudden cardiac death usually outside the hospital to be the first overt manifestation of heart disease for 40 to 60 of all cases ( Kannel 1987 deVreede-Swagemakers 1997 UK Heart Attack Study 1998) An analysis of out-of-hospital mortality from AMI in Pennsylvania in 1998 showed that in relatively rural areas where travel time to the hospital was 25 minutes or longer 72 of all AMI deaths occurred outside the hospital compared with a statewide average of 49 of AMI deaths occurring outside of the hospital (OrsquoNeill 2003) In New York State 55 of IHD deaths occur outside the hospital compared to 35 of stroke deaths (NYS DOH 2009b)

Hudson River PCB Contamination

During an approximate 30-year period ending in 1977 the General Electric (GE) plants in Hudson Falls and Fort Edward used PCBs in the manufacture of electrical capacitors and discharged large quantities of PCBs into the upper Hudson River Estimates of the total direct discharges to the river are as high as 1330000 pounds In 1983 the US EPA classified a 200shymile section of the Hudson River from Hudson Falls to New York City as a National Priority List site making it one of the largest Superfund sites in the United States In 2002 the US EPA issued a Record of Decision calling for targeted environmental dredging and removal of PCB-contaminated sediment in the upper River (USEPA 2002) The dredging began in May 2009

The Superfund environmental investigations and planned remedial actions have focused on the areas of the river near Hudson Falls and Fort Edward where the highest levels of PCBs in sediments are located The 2002 Record of Decision provides information about the human health risk assessment conducted for the Upper and Mid-Hudson and the ecological risk assessment conducted for the Upper and Lower Hudson River (USEPA 2002) Fish contamination and fish consumption are the primary concerns at the site

In 1976 fishing was banned in the upper river and in 1995 a catch and release policy replaced the ban NYS DOH continues to recommend people eat no fish from the Upper Hudson River that children under age 15 and women of childbearing age eat no fish from the entire 200 mile stretch of river below Hudson Falls and that the general population eat none of most species of fish caught between the federal dam at Troy and Catskill approximately 40 miles south of Troy The Lower Hudson remains closed to commercial fishing for striped bass and eight other species

The fishing bans and advisories were issued because ingestion of fish is the major potential pathway for exposure for the general population Many previous studies have shown relationships between consumption of fish from contaminated waters and human PCB levels Another potential pathway for exposure less frequently studied to date is inhalation of PCBs NYS DOH and ATSDR launched ldquoThe Hudson River Communities Projectrdquo in 2000 to address this issue (More information on this project is provided below)

POPS and PCBs

The category ldquopersistent organic pollutantsrdquo (POPs) includes man-made chemicals that are by definition persistent in the environment Examples of POPs are polychlorinated biphenyls

5

(PCBs) polychlorinated dibenzo-p-dioxins (PCDDs) polychorinated dibenso-furans (PCDFs) and pprsquo-diphenyldichloroethene (DDE) a breakdown product of the pesticide dichlorodiphenyltrichloroethane (DDT) These chemicals are stored in human fat tissues accumulating over an individualrsquos lifetime They also accumulate in fish and other species thereby concentrating up the food chain Levels of most POPs measured in blood serum of Americans have decreased in the last few decades An exception is polybrominated diphenyl ethers (PBDEs) levels of which have been increasing in recent years While levels of many POPs are declining a large number continue to be detectable in blood for most of the US population (USDHHS 2009)

This follow-up investigation focuses on the Hudson River where the POPs of concern are PCBs PCBs are a group of synthetic chlorinated compounds manufactured in the US through 1977 Production was halted due to concerns about the persistence of PCBs in the environment and the toxicity of PCBs shown in animal models PCBs were used primarily by the electrical industry for capacitors and transformers and were also used in hydraulic fluids fluorescent light fixtures flame retardants inks adhesives carbonless copy paper paints pesticide extenders plasticizers and wire insulators PCBs can be oily liquids or solids and some are volatile and can exist as vapor in air PCBs in the environment are generally mixtures of the many different types known as congeners (ATSDR 2000)

The variety of types of PCBs adds to the complexity of studying adverse animal and human health effects Increasingly PCB research focuses on specific PCB congeners and groups of congeners because specific types of congeners appear to have different mechanisms of action and toxicity Dioxin-like PCBs (DLPCBs) are an important subset of PCBs due to concerns about their toxicity The focus on dioxin-like PCBs is paralleled by increasing research on the group of POPs with dioxin-like properties This type of PCB and POP research assigns toxic potency values relative to dioxin to various PCBs and POPs to provide an overall toxic equivalency value to specific PCB and other POPs

From 2002 through 2007 NYS DOH and SUNY Albany School of Public Health researchers conducted the Hudson River Communities Study among residents of Hudson Falls and Fort Edward the villages where the PCB discharges from two GE electrical capacitor plants occurred The comparison community was upriver from the discharges in Glens Falls This studyrsquos evaluation of exposures included measuring PCBs in outdoor air and in blood serum The study found that PCB levels in outdoor air were statistically significantly higher in the Hudson FallsFort Edward area (072 nanograms per cubic meter) than in the upriver Glens Falls area (040 nanograms per cubic meter) but that the PCB concentrations in the Hudson FallsFort Edward area were lower than those in other communities with known PCB-contaminated sites and similar to levels reported in other locations in the northeastern United States A statistically significant reduction in the total PCB concentrations in outdoor air samples was observed at a distance of 1200 meters (34 mile) from the river from 0760 ngm3 to 0497 ngm3 (Fitzgerald et al 2007 Palmer et al 2008)

The average PCB concentrations in serum for the study and comparison populations did not differ significantly (307 ppb wet weight versus 323 ppb) The average toxic equivalency quantities (TEQ) for dioxin-like PCB congeners for the study and comparison area populations

6

also did not differ In addition the study found no detectable differences in study subjectsrsquo serum PCB levels associated with proximity to contaminated portions of the river or other PCB sources The study did find that lifetime consumption of Hudson River fish was associated with higher serum PCB levels (Fitzgerald 2007 Palmer 2008) Another recent biomonitoring study of Brooklyn Manhattan and New Jersey fisherman who consumed versus those who did not consume their catch from the lower Hudson River and New YorkNew Jersey harbor suggests eating fish from these waters is not associated with an increased body burden of PCBs or most other organochlorines This study did not address the inhalation pathway (Morland 2008)

PCBs and Cardiovascular Disease Outcomes

Research studies on adverse health effects associated with PCB exposure have focused on many types of outcomes organs and biological systems including cancer liver reproductive and developmental effects thyroid endocrine immune system and neurological effects as well as specific effects on the skin and eyes Relatively little research has focused to date on associations between PCB exposures and CVD outcomes The majority of human studies have assessed health effects in occupational groups having relatively high-level exposures among people consuming fish from contaminated waters or people exposed to relatively high levels due to accidental contamination of food (USDHHS NIH 2010)

PCBs have been shown to produce a variety of toxic effects in animals These effects include cancer allergies hypersensitivity damage to the central and peripheral nervous systems reproductive disorders and disruption of the immune system (ATSDR 2000) In contrast research to date does not tell us whether specific types of human health effects have occurred from the low-level exposures experienced by the general population Some human epidemiological studies of relatively low levels of PCDDs PCDFs and PCBs suggest possible relationships not only with neurodevelopment of infants but also diabetes and a condition called insulin resistance conditions that increase CVD risk (Longnecker 2001 Carpenter 2006 Arisawa 2005)

In 1998 a study of a population in Seveso Italy exposed to relatively pure TCDD (dioxin) from an industrial accident showed increased mortality from CVD respiratory diseases and diabetes (Bertazzi 1998) These findings suggested the possibility that other dioxin-like chemicals such as dioxin-like PCBs might increase risk for these diseases Some more recent animal studies described below suggest possible mechanisms of action for PCB effects that increase risk for CVD diabetes and hypertension Another relevant source of information is from research using the biological monitoring data from the National Health and Nutrition Examination Survey (NHANES) project to look at the association between PCB levels in blood and self-reported diabetes and heart disease The body of evidence based on analytical human epidemiological studies remains relatively sparse however

Some recent studies using NHANES biomonitoring data show associations between PCB levels and CVD (Lee 2006 Lee 2007 Ha 2007 Everett 2008) These studies are cross-sectional (comparisons at one point in time) evaluations of PCB levels measured in blood samples from individuals and self-reported CVD and diabetes a risk factor for CVD While these studies use individual-level biomonitoring and self-reported health outcome data they are considered to be

7

relatively weak in terms of providing evidence of causality because both exposure and outcome are measured at the same point in time Given the nature of NHANES data collection and reporting practices these studies are not able to control for possible geographical confounding which may occur if urban versus rural residence is associated with higher PCB levels and higher risk for CVD for example The higher PCB levels among African Americans in these data suggest urban residence may be playing a role in the PCB data

The results of animal studies suggest multiple mechanisms by which PCB exposures could affect cardiovascular risk as well as multiple types of adverse health outcomes One hypothesized mechanisms of action specifically for dioxin-like PCBs is that they may have anti-estrogenic effects which impact CVD risk factors such as serum cholesterol levels and blood pressure (Lind 2004) Several studies suggest PCBs interact with a specific receptor (AhR) and increase cellular oxidative stress causing inflammation cell injury and cardiovascular dysfunction (Hennig 2002) Similarly some studies suggest PCB-induced inflammation of fat cells leads to obesity and increased risk for coronary artery disease (atherosclerosis) (Arsenescu 2008)

3 FOLLOW-UP STUDY DATA SOURCES AND METHODS

This follow-up investigation will provide perspective on the 2004-2005 findings (Sergeev and Carpenter 2005 Shcherbatykh et al 2005) by producing results from analyses conducted at a more detailed level of geography the block group rather than ZIP code in order to better control for confounding When the term confounding is used by epidemioogists it refers to the problem that can arise when there is a factor such as income level that is not the factor of interest in the study but that is related to the risk factor being studied (residential location ndash proxy for exposure) and to the disease being studied (CVD) This type of factor is labeled a ldquoconfounderrdquo because it can confound ie interfere with or confuse the evaluation of associations among the variables of interest in the data All types of epidemiological studies face issues of confounding but ecological ie group-level analyses face additional challenges related to potential confounding because analyses can not account for some important individual differences within the groups that are the units of analysis (Morganstern 1998b Hertz-Picciotto 1998) This investigation was able to control for each individualrsquos age sex and race but not for individual-level risk factors such as income or education level

The follow-up investigationrsquos specific objectives are to

improve the ability to control for confounding by socio-economic status through the use of block groups rather than ZIP codes as the unit of analysis

compare the hospitalization rates among residents of block groups nearest the Hudson River to those among residents of block groups farther from the Hudson River

compare the findings of this follow-up study to those of the earlier studies

Study Area and Exposure Areas

The overall study area is comprised of the 12 counties with geographic areas abutting the Hudson River south of the source of PCB contamination from the General Electric facility in Hudson Falls The 12 counties are Westchester Rockland Putnam Orange Dutchess Ulster

8

Columbia Greene Rensselaer Albany Washington and Saratoga US Census boundaries are used to define the study area because census population estimates provide the data for the study area populationrsquos size and the populationrsquos age sex and race distribution Because the study period spanned 1990 through 2005 and populations change over time 1990 and 2000 Census data were used to linearly interpolate the between-year populations from 1991 through 1999 for each block group 2000 data were used without extrapolation to estimate block group populations from 2001 through 2005

Because block group boundaries change between the Census years 1990 block populations were normalized to fit 2000 block group boundaries before the estimation of sex age and race group specific populations In the digital map 2000 block group boundaries were overlaid on the 1990 block centroids to assign 1990 block population counts to the corresponding 2000 block group boundaries Then 1990 block populations were summed to fit the block group boundaries Next population counts by age group (0-14 15-24 25-34 35-44 45-54 55-64 65-74 75+) sex (male female) and race (white black asian other) were estimated for each year at the block group level This investigation includes all ages over 25 years and the study population is divided into four race categories white black or African-American Asian and ldquoOtherrdquo The Pacific Island group was combined with the Asian group and American Indian ldquootherrdquo and ldquomultiple racesrdquo categories are combined into a larger ldquootherrdquo category

Classification of the population into potential exposure categories for this study is based exclusively on location relative to the Hudson River Two types of exposure proxy categories were developed block group adjacency and block group centroid distance to the river shore For the adjacency categories block groups were assigned to the adjacent category if they directly abutted the Hudson River and all other block groups in the 12 counties were assigned to the nonshyadjacent category

For the distance categories a central point (centroid) within each block group was identified and used to assign block group centroid distance to the river shore The block group centroids were population-weighted which means the geographic distribution of households within 2000 census blocks were used to locate block group centroids based on the geographic population distribution By locating a population-weighted block centroid rather than a more straightforward distance centroid the assignment of block group distance to the river shore more accurately reflects distance to the river for the majority of block group residents Buffers of one-half mile and one mile from the Hudson River boundary were created for the assignment of each block group into one of the three distance categories less than frac12 mile (closest) frac12 to one mile (close) and farther than one mile from the river (distant)

Distance categories were selected a priori at the start of the study Because environmental sampling data were not collected along the length of the river included in this investigation we are not able to determine whether these distance categories represent actual differences in exposures As discussed previously a statistically significant reduction in the total PCB concentrations in outdoor air samples was observed at a distance of 1200 meters (34 mile) from the river from 0760 ngm3 to 0497 ngm3 in a study conducted in the Hudson Falls Fort Edward area near the source of PCBs (Fitzgerald 2007 Palmer 2007)

9

For the adjacency categories of the 2371 block groups in the study area 255 (11) were classified as adjacent to the Hudson River and 2116 (89) were classified as non-adjacent (Figure 2) For the distance categories 225 (9) were classified as closest to the river 216 (9) as close and 1930 block groups (81) as distant from the river

Comparison of Figures 1 and 2 show that block groups include smaller geographic areas than ZIP codes and that block groups are relatively smaller in area in urban versus rural areas The study area includes a total of 368 ZIP codes compared to a total of 2371 block groups ZIP code average population in the study area is approximately 8000 compared to block group average population of 1200 Comparison of Figures 2 and 3 shows that using population-weighted centroid distance to the river produces different exposure areas than adjacency Fifty-three (20) of the adjacent block groups were categorized for the distance indicator as distant because the population in the block group was concentrated in the area farther than one mile from the river On the other hand about 11 of the 2116 non-adjacent block groups were classified as closest (77) or close (162)

Hospitalization Data

Hospitalizations for ischemic heart disease (IHD) and cerebrovascular disease (stroke) were included in this study More specifically hospital discharge records that listed as the principal diagnosis either IHD (ICD-9-CM 410-414 and 4297) or cerebrovascular disease (ICD-9-CM 430-436) and subsets of each of these categories similar to the subsets evaluated in the Sergeev et al and Shcherbatykh studies were included The subsets are AMI and angina subsets of IHD and ischemic and hemorrhagic stroke which are subsets of cerebrovascular disease (stroke) (Principal diagnosis represents the illness for which the person was admitted to the hospital) Angina is an outcome not evaluated in the prior studies It was added here for comparative purposes because it represents an outcome appropriately addressed by primary outpatient rather than inpatient hospital care Angina is one of a group of outcomes included in the recently developed national ldquoPrevention Quality Indicatorsrdquo (USDHHS AHRQ 2007) as an indicator for geographic areas or populations for whom primary care is inadequate Table 3 lists the specific ICD-9-CM codes and definitions included in this investigation

While one of the previously published studies (Huang et al 2006) showed elevations in hypertension (high blood pressure) rates this outcome is rarely listed as the principal reason for hospitalization As a secondary diagnosis however hypertension is extremely common with substantial overlap with the primary diagnoses of IHD and stroke In addition hypertension hospitalizations were not as elevated in the ZIP codes adjacent to the Hudson River as the IHD and stroke outcomes As a practical matter geocoding the hypertension hospitalization data (1364000 hospitalizations) would overwhelm the resources for this project For these reasons hypertension was not included in this follow-up investigation

The source of the hospitalization data was the NYS DOH Statewide Planning and Research Cooperative System (SPARCS) established in 1979 to collect detailed records on discharges from hospitals located in New York State Persons seen in the emergency department but not admitted are not included in the SPARCS data Hospitalizations at US Veterans Administration

10

Hospitals are also not included in this dataset We identified the hospital transfers in the data and excluded them so that individuals are only counted once for the same event

The NYS DOH Data Protection Review Board for SPARCS hospitalization data approved our use of hospitalization records for this investigation and provided approximately 460000 records with a principal diagnosis of either IHD or stroke Because of the large number of cardiovascular hospitalizations and this studyrsquos requirement for geocoding each hospitalized personrsquos address we selected a 50 random sample of the hospital discharges with principal diagnoses of IHD or stroke for inclusion in the study (229948 records) Approximately one percent of these records (2737) were eliminated from the study because they contained no residential address resulting in 227211 hospitalization records 24939 of these records were eliminated after geocoding because they were not within the 12 county study area This left 202272 records remaining in the study

For geocoding to the block group level the hospitalization street addresses were modified as needed using US Postal Service standards The addresses were assigned geographic coordinates using commercially available geocoding software (MapMarker 2004) A combination of land parcel data and street centerline files were used to assign geographic coordinates No contact was made with cases parents legal guardians or next of kin of cases to determine residential locations

Once geographic coordinates were assigned to cases through address-matching the case locations were overlaid onto digital maps of the study area using a geographic information system so that the number of observed cases falling within the study area boundaries could be determined and a 2000 Census block group could be assigned to each hospitalization In order to protect confidentiality no maps of individual case locations are provided Geographic coordinates were not able to be assigned for some addresses for example rural routes or post office boxes Internet searches were conducted for addresses that were located in facilities such as nursing homes and trailer parks

While fewer than one percent of records were missing sex or age information approximately 15000 records seven percent were missing race information Evaluation of the geographic and temporal distribution of these records showed they were not randomly geographically distributed 47 of the records with no race information were from one hospital for that hospital race was missing from 100 of the records in all years except one after 1997 mid-way through the study time-frame Evaluation of other records missing race showed similar patterns of 100 of records missing race in specific years In order to prevent biased hospitalization rate estimates records missing race were assigned race randomly assuming a race distribution equal to that of the individualrsquos residential block group

Statistical Analysis

This investigation used negative binomial multivariable regression models the statistical methods used in the previous studies (Sergeev and Carpenter 2005 [IHD] Shcherbatykh et al 2005 [stroke]) The follow-up studyrsquos analyses compared hospitalization rates for populations in block groups adjacent or in close proximity to the Hudson River to rates for populations more

11

distant from the Hudson River while adjusting for the potential influence of other factors Sex age and race-specific hospitalization count data comprise the dependent variable and the statistical methods attempt to predict the counts using the sex age and race distributions from Census data as well as block group-level demographics such as median income Models were first estimated using basic Census information for age sex and race Then more fully adjusted models were estimated using information at the block group level for income education population density Hispanic ethnicity and distance to the nearest hospital

More specifically each hospitalized personrsquos block group was used to assign quartile categories of block group median household income of population with less than a college education population density of the population identified as Hispanic and distance from the block group centroid to the nearest hospital These variables were selected in advance The population density variable was selected to account for potential effects on hospitalization rates of large differences in population density from the relatively rural northern counties to the more urban southern counties in the study area of the population identified as Hispanic was included because Hispanic ethnicity may be associated with CVD risk but Hispanic ethnicity was not available in the hospitalization records Distance to the hospital was included because studies show that shorter distances increase the likelihood that individuals with AMI and stroke survive to be admitted to the hospital

As described previously the literature about risk factors for CVD shows strong associations between lower income and education and higher CVD risk Regarding the effects of income and education on hospitalization rates the literature from analytic studies specifically of hospitalizations for CVD is quite sparse Most analytic studies focus on prevalence incidence or mortality in population-based studies or cohort studies and there are not many of these An example of a study that was useful as background is a study of income level and CVD mortality that showed strong associations of lower income and higher CVD mortality for men and women with the relationship for men being linear throughout the range of income levels while for women the relationship was only strong at levels of income below the median (Rehkopf 2008)

This follow-up studyrsquos approach was to use the same set of variables for the more fully adjusted models using quartile cut points for all the sexrace groups and types of CVD diagnostic subsets The 2000 Census data were used to create the population-weighted quartile categories to be used in the statistical analyses The quartile cut points are shown below

population density(persons per square mile)

median 1999 household income

with less than a college education

with Hispanic ethnicity

Hospital distance

Quartile 1 61 - lt507 $2499-lt$40968 0 - lt31 0 - lt2 lt25 (median) Quartile 2 507 - lt102 $40968 - lt54650 31 - lt43 2 - lt5 25 - lt6 (50-75) Quartile 3 102 - lt6410 $54650 - lt72944 43 - lt55 5 - lt10 6 ndash 11 (75-90) Quartile 4 6410 - 94372 $72944 - 200001 55 - 100 10 - 86 11 ndash 45 (90-100)

SAS version 9 provided the statistical programs used to estimate hospitalization rates for exposure proxy categories and demographic categories The models provide ldquocontrast

12

estimatesrdquo ie rate ratios for one category versus a reference category of a variable and 95 confidence intervals (CI) for these rate ratios For the location categories rate ratios for the areas adjacent are compared to areas not adjacent to the river (reference) and areas closest and close to the river are compared to areas distant from the river (reference) If a rate ratio is greater than 100 the model has estimated an elevation of hospitalizations in that category of the study population compared to the reference population If the rate ratio is less than 100 then there are fewer than expected hospitalizations compared to the reference population

The magnitude of the elevation or deficit is also estimated by the rate ratio For instance if the hospitalization rate is twice as high in the adjacent population it would result in a rate ratio of 200 while a 50 higher rate would result in a rate ratio of 150 If the hospitalization rate in the comparison group was only half as big as in the reference group this would result in a rate ratio of 050 The 95 confidence interval assists with determining whether the observed differences are statistically meaningful ie statistically significant The confidence interval provides the range in which there is a 95 probability of including the true rate ratio

In addition to directly assess potential confounding we examined the associations between income and exposure proxy categories and between income and CVD hospitalization risk Because the modeling process provides rate ratios that provide comparisons among groups but not the rates themselves we also calculated the rates per population for each sex race and income category in order to directly observe the relative magnitudes of CVD hospitalization rates Finally analyses stratified by income quartile were also conducted in order to assess the potential role of confounding by income The stratified analyses consist of negative binomial regressions conducted separately for each quartile of income

4 RESULTS

Demographics Associations with Proximity to the River

Study Population Table 4 provides detailed demographic characteristics for the total study population and the populations of the adjacent versus non-adjacent and nearer versus farther from the river block groups from the 2000 Census The total population of the study area for this one year of the study time period is 2924631 persons The first row of Table 4 shows that approximately 11 percent of the study area population is classified as living in block groups adjacent to the Hudson River and 89 live in non-adjacent block groups Approximately 8 of the studyrsquos population lives in the closest block groups approximately 10 lives in close block groups and approximately 82 in distant block groups The table also shows that approximately 81 of the total study population is identified as white 9 black 3 Asian less than 01 Pacific Islander 02 American Indian 4 ldquootherrdquo and 2 ldquomultiple racesrdquo

The 2000 Census data (Table 4) also show that the populations in block groups adjacent and nonshyadjacent to the river are similar in terms of the percent of minority population but the adjacent block groups include a smaller percentage of persons who have attended college and a higher percentage of households below the poverty level Median income is $47320 for block groups adjacent versus $62668 for block groups not adjacent to the river Data for the study area from the 1990 Census show similar distributions (data not shown)

13

For the distance to the river categories there is a gradient for the three levels of distance from the river with the closest block groups on average having higher percentages of minorities smaller percentages of persons who have attended college and higher percentages of households below the poverty level (Table 4) The populations of the close and closest block groups are more similar to each other than to the populations of the distant block groups The population of the closest block groups is estimated to be 19 black close block groups 14 black and distant block groups 8 black The proportions for the ldquootherrdquo category used in the analyses which includes multiple races and American Indians are approximately 12 in both the closest block groups and in the close block groups compared to 5 in the block groups most distant from the river Median income for the closest block groups is $40748 close block groups $44962 and most distant block groups $64847

The bar chart below illustrates the correlations between US Census 2000 indicators of block group socio-economic status and block group centroid (population weighted) distance from the river for the study population

2000 Census Demographics for Study Population by Distance to River Category

black other hispanic no college below poverty

of pop ulati on

0

10

20

30

40

50

60

lt12 mile

12 to 1 mile

gt1 mile

Table 5 shows selected demographic information for the person-year estimates used in the study analyses for the 1990-2005 study period Person-year data represent individuals from Census estimates counted repeatedly for each year of the study timeframe As described in the Methods section the estimation of person years used interpolation to account for population changes from 1990-2000 and techniques for taking account of block boundary changes The person-years estimates for demographic categories (complete data not shown) show similar patterns as the 1990 Census data (data not shown) and 2000 Census data (Table 5) Table 5 for example shows nearly identical percentages of population in each of the river distance categories as Table 4

Table 5 also shows person year information for income quartiles and exposure proxy categories as well as race and exposure proxy categories that is consistent with the Table 4 data for the

14

2000 population Table 5 row percentages show for example that while approximately 11 of the total study population live in adjacent block groups approximately 19 of the population in the lowest income quartile live in adjacent block groups By design approximately 25 of the total study population is in each of the income quartiles Table 5 column percentages show however that in the adjacent block groups 41 of the population is in the lowest income quartile and only 9 are in the highest income quartile

For the block groups within frac12 mile of the river the distribution among income quartiles is even more skewed towards the lowest income quartile 61 of the population living within frac12 mile of the river is in the lowest income quartile and only 6 is in the highest income quartile Given that approximately 25 of the total study population is within each income quartile these percentages indicate that living within the block groups closest to the river is associated with more than a doubling of the likelihood of being in a block group in the lowest socio-economic status quartile 436 of those living in the block groups close to the river (12 to one mile from the river) are in block groups in the lowest income quartile While less than a doubling this indicates that living in the close block groups is associated with a 75 increase in the likelihood of being in the lowest quartile for socio-economic status

Hospitalizations Table 6 shows the process of removing hospitalization records from the study for a variety of reasons The first column includes the 202272 records that contained a residential address and were located within the 12-county study area as described in the Methods section The next five columns show the records remaining after data cleaning exclusion of transfers elimination of records with missing covariates and exclusions due to poor geocoding results Approximately 33000 records were eliminated because they were transfers from one hospital to another for the same event For approximately 1100 records geocoding was inadequate for assigning block group for imputing race as described in the Methods section 336 records were missing sex (11) or age (325) Of the 167733 records remaining 154220 (92) were able to be adequately geocoded to a street address for assignment in a specific block group While approximately eight percent of hospitalizations were eliminated due to inadequate geocoding Table 6 shows that the percentages of hospitalization records for white black Asian and ldquootherrdquo categories and the distribution among diagnostic categories change very little after these records were dropped

The final column of Table 6 shows that of the 154220 hospitalizations in the analysis 101012 IHD hospitalizations comprise the largest health outcome category and are 65 of the hospitalizations in the study AMI hospitalizations (35811) account for 35 of the IHD hospitalizations and angina (4731) accounts for five percent of the IHD hospitalizations Stroke hospitalizations (53208) comprise 34 of the studyrsquos hospitalizations Approximately 36 of the stroke hospitalizations are for ischemic stroke (19371) and approximately 11 are for hemorrhagic stroke (5995)

Table 7 shows the distributions of hospitalizations among the four income quartiles While each income quartile (by definition) contains approximately 25 of the study population for all types of hospitalizations we see that the lowest income quartile includes more than 25 of hospitalizations from a minimum of 286 (IHD) to a maximum of 356 of hospitalizations (angina) The highest income quartile includes fewer than 25 of hospitalizations for each CVD

15

category ranging from 182 for angina to 224 for hemorrhagic stroke These deviations from the 25 quartile distribution show a strong linear pattern with the largest percentage of hospitalizations coming from residents of block groups in the lowest income quartile and the smallest percentages coming from residents of the most wealthy block groups For AMI the lowest income quartilersquos share of hospitalizations is 50 higher than the highest income quartilersquos share of hospitalizations (30 vs 20) IHD and stroke show similar differences and angina shows a doubling with the lowest income quartile contributing 36 of hospitalizations compared to the highest income quartilersquos 18 share of hospitalizations

Table 7 also shows the distribution of hospitalizations among the race categories and the study population race category distribution for comparison Blacks show lower percentages of IHD and AMI than their population percentages and higher percentages of angina and stroke While full presentation of race and sex findings is beyond the scope of this report the lower than expected percentages of IHD and AMI among blacks who on average have lower socioshyeconomic status and are at greater risk for CVD hospitalizations raised questions that were addressed by calculating hospitalization rates per 1000 population for each sex and race group for each income quartile (Appendix B Table 2) Figures 4 5 and 6 show these rates for the white black and Asian categories (These rates for the ldquootherrdquo category are comparatively very high and are not shown on these graphs This issue is addressed in the Discussion section) Figures 4 5 and 6 show the consistent and strong effects of income quartile on hospitalization rates for AMI and stroke for whites blacks and Asians of both sexes The other outcomes show similar patterns (The ldquootherrdquo group showed an opposite pattern [See Discussion Section])

Demographics Summary

The strong associations of lower income with residence closer to the river (Table 5) and lower income with higher hospitalization rates (Tables 7 amp 8) create a difficult challenge for this investigation The multivariable analyses of hospitalization rates attempt to control for socioshyeconomic differences However because these income differences which strongly affect hospitalization rates are also strongly correlated with the gradations of exposure (river distance) it will be difficult to disentangle the influence of socio-economic status from the influence of potential exposures

Statistical Analyses Proximity to the River and CVD Hospitalization Risk

All the variables described above were included in the multivariable regression models and almost every quartile category for every variable showed statistically significant and expected types of effects for all the CVD hospitalization categories (Appendix B Tables 1a and 1b) The following presentation of results focuses on the rate ratios for the exposure proxy variables Table 8a (IHD) and 8b (stroke) show the regression modeling results Results are shown first for models that include only age sex and race in addition to the exposure proxy categories and second for models that also include median income education population density Hispanic ethnicity and distance to nearest hospital The tables provide numbers of hospitalizations and person-years to assist with interpretation For example for total IHD we can see that while 108 of the study population lives in adjacent block groups 122 of the hospitalizations are among people in adjacent block groups This difference of 14 percentage points is the basis for

16

the finding of a statistically significantly elevated rate ratio (risk) for IHD among people living in the adjacent block groups (RR=133 CI 112-157)

Looking across the columns of Tables 8a and 8b we can see the rate ratios for each of the outcomes for the adjacent versus non-adjacent block groups and the results are similar to those for IHD The elevated rate ratios range from 123 for angina to 136 for AMI The only rate ratio not elevated in these first models adjusting only for age sex and race is for hemorrhagic stroke Going down the rows of the Table the next set of rate ratios are for the distance to the river categories The findings are very similar to those for the adjacent block groups Here however we have two exposure proxy categories allowing comparison of the two These rate ratios show no evidence that living in the block groups closest to the river (lt12 mile to the river) produces greater risk than living in the block groups close to the river (12 to 1 mile from the river)

The next set of rate ratios are for models with additional adjustment for the demographics of each hospitalized personrsquos block group as well as distance to the nearest hospital As expected adjustment for block group median income and other demographics accounts for some of the variability in hospitalization counts and thereby reduces the magnitude of the rate ratios estimated for the river adjacency and distance categories The rate ratio elevation for total IHD for example is reduced from 133 to 107 (RR=107 CI 104-111) The elevated rate ratios in these more fully adjusted models for the adjacency analyses are of similarly small magnitude ranging from 106 for AMI to 110 for ischemic stroke Four of the six outcomes show statistically significant elevations Again only hemorrhagic stroke shows no evidence of an elevated rate ratio

For the river distance categories none of the rate ratios show any evidence that living in the block groups closest to the river (lt12 mile to the river) produces greater risk than living in the block groups close to the river (12 to 1 mile) Of the six types of outcomes only two show statistically significantly elevated rate ratios for living in the closest block groups (total stroke and ischemic stroke) For the close block groups four out of six outcomes show statistically significantly elevated rate ratios The statistically significantly elevated rate ratios vary from 106 (AMI stroke) to 113 (angina)

To address the strong evidence described earlier for confounding in these data ie the association of lower income with both CVD hospitalizations and residence closer to the river we also conducted the analyses separately for each quartile of income These results are presented in Table 9 The first results are from analyses adjusting only for age sex and race and the second set of results are from models including the additional demographic factors For simplicity rate ratios are shown only for block groups closest to the river compared as in the other models to block groups farther than one mile from the river (Results were similar for the other distance categories (data not shown)

For the two lowest income quartiles Table 9 shows results similar to those for the study population as a whole (Table 8a 8b) However for the two highest income quartiles which represent frac12 of the study population Table 9 shows a consistent pattern of reduced rate ratios for residence in the closest block groups within frac12 mile of the river In the highest income quartile

17

(Tables 8a amp 8b) are the result of the inability to completely control for socio-economic statusrsquo effects on CVD hospitalization risk The strong effect of socio-economic status on residential location (exposure proxy) and on CVD hospitalization risk makes it likely that these overall multivariable regression results are affected by uncontrolled ie residual confounding by socioshyeconomic status

Because of the similarities in study design and statistical analysis the prior studyrsquos findings can be compared to some degree with the findings of the multivariate regression analyses in the current study The major difference between the studies is the composition of the comparison areas for the analyses Because of this difference as well as other differences described in the report the studies are not entirely comparable

CVD hospitalization rate ratios estimated for living in block groups adjacent closest or close to the river in the current study are substantially lower than the rate ratios for residence in ZIP codes adjacent to the Hudson River in the previously published studies In most cases the current analyses show elevations below 10 in contrast to elevations of 20 to 39 in the prior studies The results of the current study are quite similar to the prior study results in terms of the associations reported for age sex race and income These similarities and the general methodological similarities between the current and prior studies suggest the likelihood that the prior studiesrsquo findings also were affected by confounding by socio-economic status

Conclusions and Recommendations While this type of study can neither prove nor disprove causal links among risk factors and health effects it is useful for examining evidence for unusual patterns that might warrant additional investigation This evaluation of hospitalization data residential location and socio-economic factors in block groups in the 12 Counties abutting the Hudson River revealed a closely coinciding pattern of lower socio-economic status and residential proximity to the Hudson River the proxy for potential exposure and a strong association of lower socio-economic status and risk for CVD hospitalizations The strong associations among residential location socio-economic status and CVD risk prevent us from drawing definitive conclusions from the multivariable regression analyses that showed modest but statistically significant elevations of CVD hospitalization risk for residents of block groups near the river

To draw more definitive conclusions particularly for outcomes such as CVD that are determined by multiple known risk factors studies seeking to investigate the role of PCB or POP exposures will require resource-intensive methods that include gathering individual-level exposure or biomonitoring information as well as medical histories and information on other risk factors

This current investigation did not include environmental or biological sampling data One study did gather such data in the Hudson River region with the highest potential exposures (Hudson Falls Fort Edward) This study detected a statistically significant reduction in the total PCB concentrations in outdoor air samples at a distance of 1200 meters (34 mile) from the river However the PCB levels in outdoor air in this area were lower than those in other communities with known PCB-contaminated sites and similar to levels reported in other locations in the northeastern United States The study found no detectable differences in study subjectsrsquo serum

iv

PCB levels associated with proximity to contaminated portions of the river or other PCB sources (Fitzgerald 2007 Palmer 2008)

It is recommended that the results of the current investigation of CVD hospitalization rates and residence near the Hudson River be shared with NYS residents and other stakeholders with interest in this issue and be submitted for publication in a peer-reviewed journal

v

1 INTRODUCTION

The New York State Department of Health (NYS DOH) has a cooperative agreement with the Agency for Toxic Substances and Disease Registry (ATSDR) to perform environmental exposure and health assessments conduct health statistics reviews and perform epidemiological studies of populations in New York State which may have been exposed to environmental contaminants In October 2006 Member of Congress Maurice Hinchey requested that ATSDR conduct a health consultation to assess cardiovascular disease among people residing near the Hudson River A 200-mile segment of the Hudson River from Hudson Falls south to New York City is a National Priority List (NPL) (Superfund) site due to contamination of the river with polychlorinated biphenyls (PCBs) from GE sites in Hudson Falls and Ft Edward (See Appendix A for Member of Congress Hincheyrsquos letter)

In support of his request Member of Congress Hinchey cited three studies (Sergeev and Carpenter 2005 [ischemic heart disease] Shcherbatykh et al 2005 [stroke] Huang et al 2006 [hypertension]) These statewide studies suggest residents of ZIP codes containing hazardous waste sites with persistent organic pollutants (POPs) such as PCBs have elevated risk for hospitalizations for cardiovascular disease including ischemic heart disease stroke and hypertension Analyses conducted separately for ZIP codes next to the Hudson River suggest residents of these specific ZIP codes also have increased rates of hospitalization for cardiovascular disease

These studies published in 2005-2006 are ecological studies In an ecological study data are analyzed by evaluating risk factors and health outcomes among groups For example rates of disease are compared for populations of ZIP codes with varying median income levels to see if income levels appear to be related to disease outcome rates In these previously conducted studies rates of hospitalization for residents of ZIP codes with hazardous waste sites containing POPs were compared with hospitalization rates for residents of ZIP codes without such waste sites to look for evidence that potential exposures might be impacting health outcomes The studies report statistically significant elevations of cardiovascular disease hospitalizations in ZIP codes with POPs For coronary heart disease and a subset acute myocardial infarction the study showed elevations of approximately 15 and 20 respectively (Sergeev and Carpenter 2005) for stroke a 15 elevation (Shcherbatykh et al 2005) and for hypertension a 19 elevation (Huang et al 2006) in ZIP codes containing waste sites with POPs compared with hospitalization rates in other New York State ZIP codes (excluding NYC) that did not contain waste sites with POPs (Table 1)

In sub-analyses in these studies cardiovascular hospitalization rates for ZIP codes adjacent to the Hudson River were compared with cardiovascular hospitalization rates for ZIP codes in the rest of the state (excluding NYC) identified as not having inactive hazardous waste sites with POPs These sub-analyses showed statistically significantly elevated hospitalization rates of 36 and 39 for coronary heart disease and the subset acute myocardial infarction a 20 elevation for stroke and a 14 elevation for hypertension (Table 1) The Sergeev and Carpenter Shcherbatykh et al and Huang et al articles state that the Hudson River ZIP codes are in an area where people have higher income smoke less exercise more and have healthier diets This information is cited in these articles as evidence that known risk factors related to lower socioshy

1

economic status are likely not the cause for the elevated hospitalization rates in these Hudson River area ZIP codes

In these previously published reports Behavioral Risk Factor Surveillance System (BRFSS) data for entire counties are provided to support statements about the population of Hudson River ZIP codes having higher income and healthier lifestyles However because the ZIP codes adjacent to the Hudson River include only a proportion of the population of the counties abutting the river these county-level BRFSS data may be misleading regarding characteristics of the ZIP codes adjacent to the river (Figure 1) In fact data from the US Census show that the ZIP codes adjacent to the river have lower median income than the counties within which they are located The median income for the total population of the twelve counties adjacent to the Hudson River was $61010 in 2000 while the median income for the ZIP codes adjacent to the river was $45356 This median income for the population of the ZIP codes adjacent to the river is also slightly lower than the statewide (excluding NYC) median income of $47735 (US Census 2000) Thus the Hudson River ZIP codes analyzed in the Sergeev and Carpenter Shcherbatykh et al and Huang et al studies include a population of lower not higher socio-economic status compared to the rest of NYS (excluding NYC)

Because CVD and many CVD risk factors are associated with lower socio-economic status it is especially important to adjust as much as possible for socio-economic status when conducting comparative analyses of CVD hospitalizations or disease incidence or prevalence Regarding the hypothesis about POPs and CVD research studies described in more detail below have suggested possible biological mechanisms for POP exposuresrsquo effects on the heart There is very little research on human exposures to POPs and potential cardiovascular effects however

In response to Congressman Hincheyrsquos request for follow-up of the 2004-2005 findings ATSDR and NYS DOH agreed to conduct a similar study also ecological but at a different level of geography the Census block group By using a level of geography that is generally smaller in area and population than a ZIP code this follow-up seeks to more accurately adjust for socioshyeconomic and other factors that affect cardiovascular hospitalizations and to improve upon the geographic classification of relative distance to the Hudson River In addition rather than comparing hospitalizations among residents of Hudson River area ZIP codes to hospitalizations among residents of all other NYS (exclusive of NYC) ZIP codes with no waste sites with POPs this follow-up investigationrsquos analyses are restricted to the twelve counties abutting the Hudson river from Washington County south to Westchester County Within the study arearsquos twelve counties cardiovascular hospitalization rates in areas closest to the Hudson River are compared with rates in areas farther from the river This more localized approach may also assist with reducing the potential impact of regional differences in medical care practice or reporting factors that could affect the hospitalization data (Ko et al 2007) (Table 2 shows a more detailed listing of the methods used in the prior studies and follow-up study)

Regarding environmental exposures it is important to note that living near a hazardous waste site that contains PCBs or other POPs does not necessarily mean that human exposure from the site has occurred Depending upon the conditions at a site or the specific chemicals exposures may be highly improbable Information on body burden for example PCB levels in blood or on levels of the chemical in off-site air dust or soil is needed to support a claim of exposure The

2

studies published in 2005-2006 did not provide specific information such as environmental or biomonitoring data in support of the assumption that residence in ZIP codes containing hazardous waste sites with POPs or residence in ZIP codes adjacent to the Hudson River resulted in increased exposures to POPs in the population For the purpose of this follow-up we did not conduct environmental or human biological sampling to address this issue Rather this follow-up replicates the prior studiesrsquo approach with some enhancements to better control for confounding particularly confounding by socio-economic status

This descriptive epidemiologic investigation uses already existing hospitalization data from the Statewide Planning and Research Cooperative System and population data from the United States Census to evaluate comparative rates of cardiovascular hospitalizations While this type of investigation can not show proof of cause and effect it can assist in interpreting relationships among indicators of CVD burden and hypothesized exposures associated with distance from the Hudson River This descriptive investigation may also assist in determining whether additional health investigations of these associations are warranted

2 BACKGROUND

Cardiovascular Disease

The widely known risk factors for cardiovascular disease (CVD) in addition to age sex race and family history include tobacco use high blood pressure high serum cholesterol alcohol use obesity low fruit and vegetable intake physical inactivity diabetes low socio-economic status mental ill-health psychosocial stress and the use of certain medications (WHO 2010 NYS DOH 2009 Wing 1988 NYS DOH 2009a)

CVD Definitions CVD is a general term for any disease of the circulatory system The two main types of CVD are ischemic heart disease also termed coronary heart disease and cerebrovascular disease also called stroke Ischemic heart disease (IHD) results from restricted blood flow through the arteries supplying the heart muscle The most serious risk from IHD is acute myocardial infarction (AMI) another term for heart attack AMI occurs when blood flow to the heart is suddenly reduced or stopped usually due to a blockage in a coronary artery Angina is severe constricting pain in the chest often radiating from the area immediately over the heart to the left shoulder and down the arm Cerebrovascular disease refers primarily to stroke the sudden impairment of brain function resulting from interruption of circulation to the brain Ischemic stroke results from blockage of an artery supplying blood to the brain Hemorrhagic stroke occurs due to the escape of blood from a ruptured artery supplying blood to the brain

CVD Burden and Trends CVD is the leading cause of death for men and women of all ages in the United States and in New York State In 1999 for every death from IHD there were five hospitalizations for ischemic disease in NYS (NYS DOH 2009a) Mortality rates from CVD have been declining steadily since the 1980rsquos Research studies estimate that approximately half of the decline is due to improved treatments for CVD and its risk factors such as high cholesterol and high blood pressure and approximately half is due to decreasing incidence of CVD The decrease in incidence is due to trends in health behaviors primarily reduced smoking rates

3

CVD mortality rates do not appear to be declining for all segments of the population however and may actually be increasing among some groups particularly people of lower socio-economic status (Barnett et al 1999) This may be due in part to higher rates of smoking among people of lower socio-economic status as well as to other factors related to socio-economic status and race A growing literature points to multiple causes for CVD and other health disparities including health practices psychosocial stress limited resources discrimination and access to health care (Health United States 2007) Increases in the prevalence of diabetes and obesity in the general population partially offset advances in treatment and favorable trends for other risk factors An increasing ischemic heart disease mortality rate in women ages 35 to 54 since 1997 is attributed to these unfavorable risk factor trends (Ford 2007 Lloyd-Jones 2009) New York State (NYS) and other northeast states have high IHD and low stroke mortality with NYS having the highest IHD mortality and lowest stroke mortality among the 50 states (Howard et al 2009)

CVD Data Sources CVD mortality rates are more easily measured than incidence prevalence or hospitalizations because mortality data from vital records are a comprehensive and relatively accessible data source Information for estimating the incidence and prevalence of CVD outcomes comes primarily from health interview physical examination and record surveys Hospitalization data are available for NYS and the US but are generally collected and analyzed in order to track trends in hospital care and resources expended Actual CVD incidence data validated by physician diagnosis physician examination or reviews of medical records are not directly available for the entire US or for specific geographic areas of NYS Such information is available from population-based cohort studies that follow people over time in selected communities While the concern underlying this research project is to understand CVD levels in the community ie incidence and prevalence we do not have data that directly provide us with this information In the absence of such data hospitalizations are frequently used as indicators of the burden of disease in communities particularly in surveillance or hypothesis-generating investigations such as this one In NYS the Behavioral Risk Factor Surveillance System (BRFSS) an ongoing monthly telephone survey of adults aged 18 years and older provides information about the burden of CVD by age sex race income and education

CVD Risk Factor Data The BRFSS data for NYS show that over 20 of the population aged 65 and older reported having been diagnosed with CVD versus less than five percent of the population under 65 Males were slightly more likely to report having CVD than females Nonshyhispanic whites compared to other ethnic groups combined were slightly more likely to report having CVD None of these slight differences were statistically significant however The only statistically significant demographic risk factors in the BRFSS data were low income and education levels Statistically significant elevations of reported CVD were observed for respondents with incomes below $25000 and respondents with a high school degree or less education (NYS DOH 2009a)

Another factor that can affect the outcome measure hospitalizations is the likelihood of mortality prior to reaching the hospital Studies show a variety of factors affect whether an individual experiencing a heart attack or stroke outside the hospital survives to be counted in hospitalization data These factors include lack of knowledge about personal risk or prior disease socio-economic status race marital status and distance to the hospital (Galea 2007

4

Barnett 2006 Ayala 2003) Population-based studies report sudden cardiac death usually outside the hospital to be the first overt manifestation of heart disease for 40 to 60 of all cases ( Kannel 1987 deVreede-Swagemakers 1997 UK Heart Attack Study 1998) An analysis of out-of-hospital mortality from AMI in Pennsylvania in 1998 showed that in relatively rural areas where travel time to the hospital was 25 minutes or longer 72 of all AMI deaths occurred outside the hospital compared with a statewide average of 49 of AMI deaths occurring outside of the hospital (OrsquoNeill 2003) In New York State 55 of IHD deaths occur outside the hospital compared to 35 of stroke deaths (NYS DOH 2009b)

Hudson River PCB Contamination

During an approximate 30-year period ending in 1977 the General Electric (GE) plants in Hudson Falls and Fort Edward used PCBs in the manufacture of electrical capacitors and discharged large quantities of PCBs into the upper Hudson River Estimates of the total direct discharges to the river are as high as 1330000 pounds In 1983 the US EPA classified a 200shymile section of the Hudson River from Hudson Falls to New York City as a National Priority List site making it one of the largest Superfund sites in the United States In 2002 the US EPA issued a Record of Decision calling for targeted environmental dredging and removal of PCB-contaminated sediment in the upper River (USEPA 2002) The dredging began in May 2009

The Superfund environmental investigations and planned remedial actions have focused on the areas of the river near Hudson Falls and Fort Edward where the highest levels of PCBs in sediments are located The 2002 Record of Decision provides information about the human health risk assessment conducted for the Upper and Mid-Hudson and the ecological risk assessment conducted for the Upper and Lower Hudson River (USEPA 2002) Fish contamination and fish consumption are the primary concerns at the site

In 1976 fishing was banned in the upper river and in 1995 a catch and release policy replaced the ban NYS DOH continues to recommend people eat no fish from the Upper Hudson River that children under age 15 and women of childbearing age eat no fish from the entire 200 mile stretch of river below Hudson Falls and that the general population eat none of most species of fish caught between the federal dam at Troy and Catskill approximately 40 miles south of Troy The Lower Hudson remains closed to commercial fishing for striped bass and eight other species

The fishing bans and advisories were issued because ingestion of fish is the major potential pathway for exposure for the general population Many previous studies have shown relationships between consumption of fish from contaminated waters and human PCB levels Another potential pathway for exposure less frequently studied to date is inhalation of PCBs NYS DOH and ATSDR launched ldquoThe Hudson River Communities Projectrdquo in 2000 to address this issue (More information on this project is provided below)

POPS and PCBs

The category ldquopersistent organic pollutantsrdquo (POPs) includes man-made chemicals that are by definition persistent in the environment Examples of POPs are polychlorinated biphenyls

5

(PCBs) polychlorinated dibenzo-p-dioxins (PCDDs) polychorinated dibenso-furans (PCDFs) and pprsquo-diphenyldichloroethene (DDE) a breakdown product of the pesticide dichlorodiphenyltrichloroethane (DDT) These chemicals are stored in human fat tissues accumulating over an individualrsquos lifetime They also accumulate in fish and other species thereby concentrating up the food chain Levels of most POPs measured in blood serum of Americans have decreased in the last few decades An exception is polybrominated diphenyl ethers (PBDEs) levels of which have been increasing in recent years While levels of many POPs are declining a large number continue to be detectable in blood for most of the US population (USDHHS 2009)

This follow-up investigation focuses on the Hudson River where the POPs of concern are PCBs PCBs are a group of synthetic chlorinated compounds manufactured in the US through 1977 Production was halted due to concerns about the persistence of PCBs in the environment and the toxicity of PCBs shown in animal models PCBs were used primarily by the electrical industry for capacitors and transformers and were also used in hydraulic fluids fluorescent light fixtures flame retardants inks adhesives carbonless copy paper paints pesticide extenders plasticizers and wire insulators PCBs can be oily liquids or solids and some are volatile and can exist as vapor in air PCBs in the environment are generally mixtures of the many different types known as congeners (ATSDR 2000)

The variety of types of PCBs adds to the complexity of studying adverse animal and human health effects Increasingly PCB research focuses on specific PCB congeners and groups of congeners because specific types of congeners appear to have different mechanisms of action and toxicity Dioxin-like PCBs (DLPCBs) are an important subset of PCBs due to concerns about their toxicity The focus on dioxin-like PCBs is paralleled by increasing research on the group of POPs with dioxin-like properties This type of PCB and POP research assigns toxic potency values relative to dioxin to various PCBs and POPs to provide an overall toxic equivalency value to specific PCB and other POPs

From 2002 through 2007 NYS DOH and SUNY Albany School of Public Health researchers conducted the Hudson River Communities Study among residents of Hudson Falls and Fort Edward the villages where the PCB discharges from two GE electrical capacitor plants occurred The comparison community was upriver from the discharges in Glens Falls This studyrsquos evaluation of exposures included measuring PCBs in outdoor air and in blood serum The study found that PCB levels in outdoor air were statistically significantly higher in the Hudson FallsFort Edward area (072 nanograms per cubic meter) than in the upriver Glens Falls area (040 nanograms per cubic meter) but that the PCB concentrations in the Hudson FallsFort Edward area were lower than those in other communities with known PCB-contaminated sites and similar to levels reported in other locations in the northeastern United States A statistically significant reduction in the total PCB concentrations in outdoor air samples was observed at a distance of 1200 meters (34 mile) from the river from 0760 ngm3 to 0497 ngm3 (Fitzgerald et al 2007 Palmer et al 2008)

The average PCB concentrations in serum for the study and comparison populations did not differ significantly (307 ppb wet weight versus 323 ppb) The average toxic equivalency quantities (TEQ) for dioxin-like PCB congeners for the study and comparison area populations

6

also did not differ In addition the study found no detectable differences in study subjectsrsquo serum PCB levels associated with proximity to contaminated portions of the river or other PCB sources The study did find that lifetime consumption of Hudson River fish was associated with higher serum PCB levels (Fitzgerald 2007 Palmer 2008) Another recent biomonitoring study of Brooklyn Manhattan and New Jersey fisherman who consumed versus those who did not consume their catch from the lower Hudson River and New YorkNew Jersey harbor suggests eating fish from these waters is not associated with an increased body burden of PCBs or most other organochlorines This study did not address the inhalation pathway (Morland 2008)

PCBs and Cardiovascular Disease Outcomes

Research studies on adverse health effects associated with PCB exposure have focused on many types of outcomes organs and biological systems including cancer liver reproductive and developmental effects thyroid endocrine immune system and neurological effects as well as specific effects on the skin and eyes Relatively little research has focused to date on associations between PCB exposures and CVD outcomes The majority of human studies have assessed health effects in occupational groups having relatively high-level exposures among people consuming fish from contaminated waters or people exposed to relatively high levels due to accidental contamination of food (USDHHS NIH 2010)

PCBs have been shown to produce a variety of toxic effects in animals These effects include cancer allergies hypersensitivity damage to the central and peripheral nervous systems reproductive disorders and disruption of the immune system (ATSDR 2000) In contrast research to date does not tell us whether specific types of human health effects have occurred from the low-level exposures experienced by the general population Some human epidemiological studies of relatively low levels of PCDDs PCDFs and PCBs suggest possible relationships not only with neurodevelopment of infants but also diabetes and a condition called insulin resistance conditions that increase CVD risk (Longnecker 2001 Carpenter 2006 Arisawa 2005)

In 1998 a study of a population in Seveso Italy exposed to relatively pure TCDD (dioxin) from an industrial accident showed increased mortality from CVD respiratory diseases and diabetes (Bertazzi 1998) These findings suggested the possibility that other dioxin-like chemicals such as dioxin-like PCBs might increase risk for these diseases Some more recent animal studies described below suggest possible mechanisms of action for PCB effects that increase risk for CVD diabetes and hypertension Another relevant source of information is from research using the biological monitoring data from the National Health and Nutrition Examination Survey (NHANES) project to look at the association between PCB levels in blood and self-reported diabetes and heart disease The body of evidence based on analytical human epidemiological studies remains relatively sparse however

Some recent studies using NHANES biomonitoring data show associations between PCB levels and CVD (Lee 2006 Lee 2007 Ha 2007 Everett 2008) These studies are cross-sectional (comparisons at one point in time) evaluations of PCB levels measured in blood samples from individuals and self-reported CVD and diabetes a risk factor for CVD While these studies use individual-level biomonitoring and self-reported health outcome data they are considered to be

7

relatively weak in terms of providing evidence of causality because both exposure and outcome are measured at the same point in time Given the nature of NHANES data collection and reporting practices these studies are not able to control for possible geographical confounding which may occur if urban versus rural residence is associated with higher PCB levels and higher risk for CVD for example The higher PCB levels among African Americans in these data suggest urban residence may be playing a role in the PCB data

The results of animal studies suggest multiple mechanisms by which PCB exposures could affect cardiovascular risk as well as multiple types of adverse health outcomes One hypothesized mechanisms of action specifically for dioxin-like PCBs is that they may have anti-estrogenic effects which impact CVD risk factors such as serum cholesterol levels and blood pressure (Lind 2004) Several studies suggest PCBs interact with a specific receptor (AhR) and increase cellular oxidative stress causing inflammation cell injury and cardiovascular dysfunction (Hennig 2002) Similarly some studies suggest PCB-induced inflammation of fat cells leads to obesity and increased risk for coronary artery disease (atherosclerosis) (Arsenescu 2008)

3 FOLLOW-UP STUDY DATA SOURCES AND METHODS

This follow-up investigation will provide perspective on the 2004-2005 findings (Sergeev and Carpenter 2005 Shcherbatykh et al 2005) by producing results from analyses conducted at a more detailed level of geography the block group rather than ZIP code in order to better control for confounding When the term confounding is used by epidemioogists it refers to the problem that can arise when there is a factor such as income level that is not the factor of interest in the study but that is related to the risk factor being studied (residential location ndash proxy for exposure) and to the disease being studied (CVD) This type of factor is labeled a ldquoconfounderrdquo because it can confound ie interfere with or confuse the evaluation of associations among the variables of interest in the data All types of epidemiological studies face issues of confounding but ecological ie group-level analyses face additional challenges related to potential confounding because analyses can not account for some important individual differences within the groups that are the units of analysis (Morganstern 1998b Hertz-Picciotto 1998) This investigation was able to control for each individualrsquos age sex and race but not for individual-level risk factors such as income or education level

The follow-up investigationrsquos specific objectives are to

improve the ability to control for confounding by socio-economic status through the use of block groups rather than ZIP codes as the unit of analysis

compare the hospitalization rates among residents of block groups nearest the Hudson River to those among residents of block groups farther from the Hudson River

compare the findings of this follow-up study to those of the earlier studies

Study Area and Exposure Areas

The overall study area is comprised of the 12 counties with geographic areas abutting the Hudson River south of the source of PCB contamination from the General Electric facility in Hudson Falls The 12 counties are Westchester Rockland Putnam Orange Dutchess Ulster

8

Columbia Greene Rensselaer Albany Washington and Saratoga US Census boundaries are used to define the study area because census population estimates provide the data for the study area populationrsquos size and the populationrsquos age sex and race distribution Because the study period spanned 1990 through 2005 and populations change over time 1990 and 2000 Census data were used to linearly interpolate the between-year populations from 1991 through 1999 for each block group 2000 data were used without extrapolation to estimate block group populations from 2001 through 2005

Because block group boundaries change between the Census years 1990 block populations were normalized to fit 2000 block group boundaries before the estimation of sex age and race group specific populations In the digital map 2000 block group boundaries were overlaid on the 1990 block centroids to assign 1990 block population counts to the corresponding 2000 block group boundaries Then 1990 block populations were summed to fit the block group boundaries Next population counts by age group (0-14 15-24 25-34 35-44 45-54 55-64 65-74 75+) sex (male female) and race (white black asian other) were estimated for each year at the block group level This investigation includes all ages over 25 years and the study population is divided into four race categories white black or African-American Asian and ldquoOtherrdquo The Pacific Island group was combined with the Asian group and American Indian ldquootherrdquo and ldquomultiple racesrdquo categories are combined into a larger ldquootherrdquo category

Classification of the population into potential exposure categories for this study is based exclusively on location relative to the Hudson River Two types of exposure proxy categories were developed block group adjacency and block group centroid distance to the river shore For the adjacency categories block groups were assigned to the adjacent category if they directly abutted the Hudson River and all other block groups in the 12 counties were assigned to the nonshyadjacent category

For the distance categories a central point (centroid) within each block group was identified and used to assign block group centroid distance to the river shore The block group centroids were population-weighted which means the geographic distribution of households within 2000 census blocks were used to locate block group centroids based on the geographic population distribution By locating a population-weighted block centroid rather than a more straightforward distance centroid the assignment of block group distance to the river shore more accurately reflects distance to the river for the majority of block group residents Buffers of one-half mile and one mile from the Hudson River boundary were created for the assignment of each block group into one of the three distance categories less than frac12 mile (closest) frac12 to one mile (close) and farther than one mile from the river (distant)

Distance categories were selected a priori at the start of the study Because environmental sampling data were not collected along the length of the river included in this investigation we are not able to determine whether these distance categories represent actual differences in exposures As discussed previously a statistically significant reduction in the total PCB concentrations in outdoor air samples was observed at a distance of 1200 meters (34 mile) from the river from 0760 ngm3 to 0497 ngm3 in a study conducted in the Hudson Falls Fort Edward area near the source of PCBs (Fitzgerald 2007 Palmer 2007)

9

For the adjacency categories of the 2371 block groups in the study area 255 (11) were classified as adjacent to the Hudson River and 2116 (89) were classified as non-adjacent (Figure 2) For the distance categories 225 (9) were classified as closest to the river 216 (9) as close and 1930 block groups (81) as distant from the river

Comparison of Figures 1 and 2 show that block groups include smaller geographic areas than ZIP codes and that block groups are relatively smaller in area in urban versus rural areas The study area includes a total of 368 ZIP codes compared to a total of 2371 block groups ZIP code average population in the study area is approximately 8000 compared to block group average population of 1200 Comparison of Figures 2 and 3 shows that using population-weighted centroid distance to the river produces different exposure areas than adjacency Fifty-three (20) of the adjacent block groups were categorized for the distance indicator as distant because the population in the block group was concentrated in the area farther than one mile from the river On the other hand about 11 of the 2116 non-adjacent block groups were classified as closest (77) or close (162)

Hospitalization Data

Hospitalizations for ischemic heart disease (IHD) and cerebrovascular disease (stroke) were included in this study More specifically hospital discharge records that listed as the principal diagnosis either IHD (ICD-9-CM 410-414 and 4297) or cerebrovascular disease (ICD-9-CM 430-436) and subsets of each of these categories similar to the subsets evaluated in the Sergeev et al and Shcherbatykh studies were included The subsets are AMI and angina subsets of IHD and ischemic and hemorrhagic stroke which are subsets of cerebrovascular disease (stroke) (Principal diagnosis represents the illness for which the person was admitted to the hospital) Angina is an outcome not evaluated in the prior studies It was added here for comparative purposes because it represents an outcome appropriately addressed by primary outpatient rather than inpatient hospital care Angina is one of a group of outcomes included in the recently developed national ldquoPrevention Quality Indicatorsrdquo (USDHHS AHRQ 2007) as an indicator for geographic areas or populations for whom primary care is inadequate Table 3 lists the specific ICD-9-CM codes and definitions included in this investigation

While one of the previously published studies (Huang et al 2006) showed elevations in hypertension (high blood pressure) rates this outcome is rarely listed as the principal reason for hospitalization As a secondary diagnosis however hypertension is extremely common with substantial overlap with the primary diagnoses of IHD and stroke In addition hypertension hospitalizations were not as elevated in the ZIP codes adjacent to the Hudson River as the IHD and stroke outcomes As a practical matter geocoding the hypertension hospitalization data (1364000 hospitalizations) would overwhelm the resources for this project For these reasons hypertension was not included in this follow-up investigation

The source of the hospitalization data was the NYS DOH Statewide Planning and Research Cooperative System (SPARCS) established in 1979 to collect detailed records on discharges from hospitals located in New York State Persons seen in the emergency department but not admitted are not included in the SPARCS data Hospitalizations at US Veterans Administration

10

Hospitals are also not included in this dataset We identified the hospital transfers in the data and excluded them so that individuals are only counted once for the same event

The NYS DOH Data Protection Review Board for SPARCS hospitalization data approved our use of hospitalization records for this investigation and provided approximately 460000 records with a principal diagnosis of either IHD or stroke Because of the large number of cardiovascular hospitalizations and this studyrsquos requirement for geocoding each hospitalized personrsquos address we selected a 50 random sample of the hospital discharges with principal diagnoses of IHD or stroke for inclusion in the study (229948 records) Approximately one percent of these records (2737) were eliminated from the study because they contained no residential address resulting in 227211 hospitalization records 24939 of these records were eliminated after geocoding because they were not within the 12 county study area This left 202272 records remaining in the study

For geocoding to the block group level the hospitalization street addresses were modified as needed using US Postal Service standards The addresses were assigned geographic coordinates using commercially available geocoding software (MapMarker 2004) A combination of land parcel data and street centerline files were used to assign geographic coordinates No contact was made with cases parents legal guardians or next of kin of cases to determine residential locations

Once geographic coordinates were assigned to cases through address-matching the case locations were overlaid onto digital maps of the study area using a geographic information system so that the number of observed cases falling within the study area boundaries could be determined and a 2000 Census block group could be assigned to each hospitalization In order to protect confidentiality no maps of individual case locations are provided Geographic coordinates were not able to be assigned for some addresses for example rural routes or post office boxes Internet searches were conducted for addresses that were located in facilities such as nursing homes and trailer parks

While fewer than one percent of records were missing sex or age information approximately 15000 records seven percent were missing race information Evaluation of the geographic and temporal distribution of these records showed they were not randomly geographically distributed 47 of the records with no race information were from one hospital for that hospital race was missing from 100 of the records in all years except one after 1997 mid-way through the study time-frame Evaluation of other records missing race showed similar patterns of 100 of records missing race in specific years In order to prevent biased hospitalization rate estimates records missing race were assigned race randomly assuming a race distribution equal to that of the individualrsquos residential block group

Statistical Analysis

This investigation used negative binomial multivariable regression models the statistical methods used in the previous studies (Sergeev and Carpenter 2005 [IHD] Shcherbatykh et al 2005 [stroke]) The follow-up studyrsquos analyses compared hospitalization rates for populations in block groups adjacent or in close proximity to the Hudson River to rates for populations more

11

distant from the Hudson River while adjusting for the potential influence of other factors Sex age and race-specific hospitalization count data comprise the dependent variable and the statistical methods attempt to predict the counts using the sex age and race distributions from Census data as well as block group-level demographics such as median income Models were first estimated using basic Census information for age sex and race Then more fully adjusted models were estimated using information at the block group level for income education population density Hispanic ethnicity and distance to the nearest hospital

More specifically each hospitalized personrsquos block group was used to assign quartile categories of block group median household income of population with less than a college education population density of the population identified as Hispanic and distance from the block group centroid to the nearest hospital These variables were selected in advance The population density variable was selected to account for potential effects on hospitalization rates of large differences in population density from the relatively rural northern counties to the more urban southern counties in the study area of the population identified as Hispanic was included because Hispanic ethnicity may be associated with CVD risk but Hispanic ethnicity was not available in the hospitalization records Distance to the hospital was included because studies show that shorter distances increase the likelihood that individuals with AMI and stroke survive to be admitted to the hospital

As described previously the literature about risk factors for CVD shows strong associations between lower income and education and higher CVD risk Regarding the effects of income and education on hospitalization rates the literature from analytic studies specifically of hospitalizations for CVD is quite sparse Most analytic studies focus on prevalence incidence or mortality in population-based studies or cohort studies and there are not many of these An example of a study that was useful as background is a study of income level and CVD mortality that showed strong associations of lower income and higher CVD mortality for men and women with the relationship for men being linear throughout the range of income levels while for women the relationship was only strong at levels of income below the median (Rehkopf 2008)

This follow-up studyrsquos approach was to use the same set of variables for the more fully adjusted models using quartile cut points for all the sexrace groups and types of CVD diagnostic subsets The 2000 Census data were used to create the population-weighted quartile categories to be used in the statistical analyses The quartile cut points are shown below

population density(persons per square mile)

median 1999 household income

with less than a college education

with Hispanic ethnicity

Hospital distance

Quartile 1 61 - lt507 $2499-lt$40968 0 - lt31 0 - lt2 lt25 (median) Quartile 2 507 - lt102 $40968 - lt54650 31 - lt43 2 - lt5 25 - lt6 (50-75) Quartile 3 102 - lt6410 $54650 - lt72944 43 - lt55 5 - lt10 6 ndash 11 (75-90) Quartile 4 6410 - 94372 $72944 - 200001 55 - 100 10 - 86 11 ndash 45 (90-100)

SAS version 9 provided the statistical programs used to estimate hospitalization rates for exposure proxy categories and demographic categories The models provide ldquocontrast

12

estimatesrdquo ie rate ratios for one category versus a reference category of a variable and 95 confidence intervals (CI) for these rate ratios For the location categories rate ratios for the areas adjacent are compared to areas not adjacent to the river (reference) and areas closest and close to the river are compared to areas distant from the river (reference) If a rate ratio is greater than 100 the model has estimated an elevation of hospitalizations in that category of the study population compared to the reference population If the rate ratio is less than 100 then there are fewer than expected hospitalizations compared to the reference population

The magnitude of the elevation or deficit is also estimated by the rate ratio For instance if the hospitalization rate is twice as high in the adjacent population it would result in a rate ratio of 200 while a 50 higher rate would result in a rate ratio of 150 If the hospitalization rate in the comparison group was only half as big as in the reference group this would result in a rate ratio of 050 The 95 confidence interval assists with determining whether the observed differences are statistically meaningful ie statistically significant The confidence interval provides the range in which there is a 95 probability of including the true rate ratio

In addition to directly assess potential confounding we examined the associations between income and exposure proxy categories and between income and CVD hospitalization risk Because the modeling process provides rate ratios that provide comparisons among groups but not the rates themselves we also calculated the rates per population for each sex race and income category in order to directly observe the relative magnitudes of CVD hospitalization rates Finally analyses stratified by income quartile were also conducted in order to assess the potential role of confounding by income The stratified analyses consist of negative binomial regressions conducted separately for each quartile of income

4 RESULTS

Demographics Associations with Proximity to the River

Study Population Table 4 provides detailed demographic characteristics for the total study population and the populations of the adjacent versus non-adjacent and nearer versus farther from the river block groups from the 2000 Census The total population of the study area for this one year of the study time period is 2924631 persons The first row of Table 4 shows that approximately 11 percent of the study area population is classified as living in block groups adjacent to the Hudson River and 89 live in non-adjacent block groups Approximately 8 of the studyrsquos population lives in the closest block groups approximately 10 lives in close block groups and approximately 82 in distant block groups The table also shows that approximately 81 of the total study population is identified as white 9 black 3 Asian less than 01 Pacific Islander 02 American Indian 4 ldquootherrdquo and 2 ldquomultiple racesrdquo

The 2000 Census data (Table 4) also show that the populations in block groups adjacent and nonshyadjacent to the river are similar in terms of the percent of minority population but the adjacent block groups include a smaller percentage of persons who have attended college and a higher percentage of households below the poverty level Median income is $47320 for block groups adjacent versus $62668 for block groups not adjacent to the river Data for the study area from the 1990 Census show similar distributions (data not shown)

13

For the distance to the river categories there is a gradient for the three levels of distance from the river with the closest block groups on average having higher percentages of minorities smaller percentages of persons who have attended college and higher percentages of households below the poverty level (Table 4) The populations of the close and closest block groups are more similar to each other than to the populations of the distant block groups The population of the closest block groups is estimated to be 19 black close block groups 14 black and distant block groups 8 black The proportions for the ldquootherrdquo category used in the analyses which includes multiple races and American Indians are approximately 12 in both the closest block groups and in the close block groups compared to 5 in the block groups most distant from the river Median income for the closest block groups is $40748 close block groups $44962 and most distant block groups $64847

The bar chart below illustrates the correlations between US Census 2000 indicators of block group socio-economic status and block group centroid (population weighted) distance from the river for the study population

2000 Census Demographics for Study Population by Distance to River Category

black other hispanic no college below poverty

of pop ulati on

0

10

20

30

40

50

60

lt12 mile

12 to 1 mile

gt1 mile

Table 5 shows selected demographic information for the person-year estimates used in the study analyses for the 1990-2005 study period Person-year data represent individuals from Census estimates counted repeatedly for each year of the study timeframe As described in the Methods section the estimation of person years used interpolation to account for population changes from 1990-2000 and techniques for taking account of block boundary changes The person-years estimates for demographic categories (complete data not shown) show similar patterns as the 1990 Census data (data not shown) and 2000 Census data (Table 5) Table 5 for example shows nearly identical percentages of population in each of the river distance categories as Table 4

Table 5 also shows person year information for income quartiles and exposure proxy categories as well as race and exposure proxy categories that is consistent with the Table 4 data for the

14

2000 population Table 5 row percentages show for example that while approximately 11 of the total study population live in adjacent block groups approximately 19 of the population in the lowest income quartile live in adjacent block groups By design approximately 25 of the total study population is in each of the income quartiles Table 5 column percentages show however that in the adjacent block groups 41 of the population is in the lowest income quartile and only 9 are in the highest income quartile

For the block groups within frac12 mile of the river the distribution among income quartiles is even more skewed towards the lowest income quartile 61 of the population living within frac12 mile of the river is in the lowest income quartile and only 6 is in the highest income quartile Given that approximately 25 of the total study population is within each income quartile these percentages indicate that living within the block groups closest to the river is associated with more than a doubling of the likelihood of being in a block group in the lowest socio-economic status quartile 436 of those living in the block groups close to the river (12 to one mile from the river) are in block groups in the lowest income quartile While less than a doubling this indicates that living in the close block groups is associated with a 75 increase in the likelihood of being in the lowest quartile for socio-economic status

Hospitalizations Table 6 shows the process of removing hospitalization records from the study for a variety of reasons The first column includes the 202272 records that contained a residential address and were located within the 12-county study area as described in the Methods section The next five columns show the records remaining after data cleaning exclusion of transfers elimination of records with missing covariates and exclusions due to poor geocoding results Approximately 33000 records were eliminated because they were transfers from one hospital to another for the same event For approximately 1100 records geocoding was inadequate for assigning block group for imputing race as described in the Methods section 336 records were missing sex (11) or age (325) Of the 167733 records remaining 154220 (92) were able to be adequately geocoded to a street address for assignment in a specific block group While approximately eight percent of hospitalizations were eliminated due to inadequate geocoding Table 6 shows that the percentages of hospitalization records for white black Asian and ldquootherrdquo categories and the distribution among diagnostic categories change very little after these records were dropped

The final column of Table 6 shows that of the 154220 hospitalizations in the analysis 101012 IHD hospitalizations comprise the largest health outcome category and are 65 of the hospitalizations in the study AMI hospitalizations (35811) account for 35 of the IHD hospitalizations and angina (4731) accounts for five percent of the IHD hospitalizations Stroke hospitalizations (53208) comprise 34 of the studyrsquos hospitalizations Approximately 36 of the stroke hospitalizations are for ischemic stroke (19371) and approximately 11 are for hemorrhagic stroke (5995)

Table 7 shows the distributions of hospitalizations among the four income quartiles While each income quartile (by definition) contains approximately 25 of the study population for all types of hospitalizations we see that the lowest income quartile includes more than 25 of hospitalizations from a minimum of 286 (IHD) to a maximum of 356 of hospitalizations (angina) The highest income quartile includes fewer than 25 of hospitalizations for each CVD

15

category ranging from 182 for angina to 224 for hemorrhagic stroke These deviations from the 25 quartile distribution show a strong linear pattern with the largest percentage of hospitalizations coming from residents of block groups in the lowest income quartile and the smallest percentages coming from residents of the most wealthy block groups For AMI the lowest income quartilersquos share of hospitalizations is 50 higher than the highest income quartilersquos share of hospitalizations (30 vs 20) IHD and stroke show similar differences and angina shows a doubling with the lowest income quartile contributing 36 of hospitalizations compared to the highest income quartilersquos 18 share of hospitalizations

Table 7 also shows the distribution of hospitalizations among the race categories and the study population race category distribution for comparison Blacks show lower percentages of IHD and AMI than their population percentages and higher percentages of angina and stroke While full presentation of race and sex findings is beyond the scope of this report the lower than expected percentages of IHD and AMI among blacks who on average have lower socioshyeconomic status and are at greater risk for CVD hospitalizations raised questions that were addressed by calculating hospitalization rates per 1000 population for each sex and race group for each income quartile (Appendix B Table 2) Figures 4 5 and 6 show these rates for the white black and Asian categories (These rates for the ldquootherrdquo category are comparatively very high and are not shown on these graphs This issue is addressed in the Discussion section) Figures 4 5 and 6 show the consistent and strong effects of income quartile on hospitalization rates for AMI and stroke for whites blacks and Asians of both sexes The other outcomes show similar patterns (The ldquootherrdquo group showed an opposite pattern [See Discussion Section])

Demographics Summary

The strong associations of lower income with residence closer to the river (Table 5) and lower income with higher hospitalization rates (Tables 7 amp 8) create a difficult challenge for this investigation The multivariable analyses of hospitalization rates attempt to control for socioshyeconomic differences However because these income differences which strongly affect hospitalization rates are also strongly correlated with the gradations of exposure (river distance) it will be difficult to disentangle the influence of socio-economic status from the influence of potential exposures

Statistical Analyses Proximity to the River and CVD Hospitalization Risk

All the variables described above were included in the multivariable regression models and almost every quartile category for every variable showed statistically significant and expected types of effects for all the CVD hospitalization categories (Appendix B Tables 1a and 1b) The following presentation of results focuses on the rate ratios for the exposure proxy variables Table 8a (IHD) and 8b (stroke) show the regression modeling results Results are shown first for models that include only age sex and race in addition to the exposure proxy categories and second for models that also include median income education population density Hispanic ethnicity and distance to nearest hospital The tables provide numbers of hospitalizations and person-years to assist with interpretation For example for total IHD we can see that while 108 of the study population lives in adjacent block groups 122 of the hospitalizations are among people in adjacent block groups This difference of 14 percentage points is the basis for

16

the finding of a statistically significantly elevated rate ratio (risk) for IHD among people living in the adjacent block groups (RR=133 CI 112-157)

Looking across the columns of Tables 8a and 8b we can see the rate ratios for each of the outcomes for the adjacent versus non-adjacent block groups and the results are similar to those for IHD The elevated rate ratios range from 123 for angina to 136 for AMI The only rate ratio not elevated in these first models adjusting only for age sex and race is for hemorrhagic stroke Going down the rows of the Table the next set of rate ratios are for the distance to the river categories The findings are very similar to those for the adjacent block groups Here however we have two exposure proxy categories allowing comparison of the two These rate ratios show no evidence that living in the block groups closest to the river (lt12 mile to the river) produces greater risk than living in the block groups close to the river (12 to 1 mile from the river)

The next set of rate ratios are for models with additional adjustment for the demographics of each hospitalized personrsquos block group as well as distance to the nearest hospital As expected adjustment for block group median income and other demographics accounts for some of the variability in hospitalization counts and thereby reduces the magnitude of the rate ratios estimated for the river adjacency and distance categories The rate ratio elevation for total IHD for example is reduced from 133 to 107 (RR=107 CI 104-111) The elevated rate ratios in these more fully adjusted models for the adjacency analyses are of similarly small magnitude ranging from 106 for AMI to 110 for ischemic stroke Four of the six outcomes show statistically significant elevations Again only hemorrhagic stroke shows no evidence of an elevated rate ratio

For the river distance categories none of the rate ratios show any evidence that living in the block groups closest to the river (lt12 mile to the river) produces greater risk than living in the block groups close to the river (12 to 1 mile) Of the six types of outcomes only two show statistically significantly elevated rate ratios for living in the closest block groups (total stroke and ischemic stroke) For the close block groups four out of six outcomes show statistically significantly elevated rate ratios The statistically significantly elevated rate ratios vary from 106 (AMI stroke) to 113 (angina)

To address the strong evidence described earlier for confounding in these data ie the association of lower income with both CVD hospitalizations and residence closer to the river we also conducted the analyses separately for each quartile of income These results are presented in Table 9 The first results are from analyses adjusting only for age sex and race and the second set of results are from models including the additional demographic factors For simplicity rate ratios are shown only for block groups closest to the river compared as in the other models to block groups farther than one mile from the river (Results were similar for the other distance categories (data not shown)

For the two lowest income quartiles Table 9 shows results similar to those for the study population as a whole (Table 8a 8b) However for the two highest income quartiles which represent frac12 of the study population Table 9 shows a consistent pattern of reduced rate ratios for residence in the closest block groups within frac12 mile of the river In the highest income quartile

17

PCB levels associated with proximity to contaminated portions of the river or other PCB sources (Fitzgerald 2007 Palmer 2008)

It is recommended that the results of the current investigation of CVD hospitalization rates and residence near the Hudson River be shared with NYS residents and other stakeholders with interest in this issue and be submitted for publication in a peer-reviewed journal

v

1 INTRODUCTION

The New York State Department of Health (NYS DOH) has a cooperative agreement with the Agency for Toxic Substances and Disease Registry (ATSDR) to perform environmental exposure and health assessments conduct health statistics reviews and perform epidemiological studies of populations in New York State which may have been exposed to environmental contaminants In October 2006 Member of Congress Maurice Hinchey requested that ATSDR conduct a health consultation to assess cardiovascular disease among people residing near the Hudson River A 200-mile segment of the Hudson River from Hudson Falls south to New York City is a National Priority List (NPL) (Superfund) site due to contamination of the river with polychlorinated biphenyls (PCBs) from GE sites in Hudson Falls and Ft Edward (See Appendix A for Member of Congress Hincheyrsquos letter)

In support of his request Member of Congress Hinchey cited three studies (Sergeev and Carpenter 2005 [ischemic heart disease] Shcherbatykh et al 2005 [stroke] Huang et al 2006 [hypertension]) These statewide studies suggest residents of ZIP codes containing hazardous waste sites with persistent organic pollutants (POPs) such as PCBs have elevated risk for hospitalizations for cardiovascular disease including ischemic heart disease stroke and hypertension Analyses conducted separately for ZIP codes next to the Hudson River suggest residents of these specific ZIP codes also have increased rates of hospitalization for cardiovascular disease

These studies published in 2005-2006 are ecological studies In an ecological study data are analyzed by evaluating risk factors and health outcomes among groups For example rates of disease are compared for populations of ZIP codes with varying median income levels to see if income levels appear to be related to disease outcome rates In these previously conducted studies rates of hospitalization for residents of ZIP codes with hazardous waste sites containing POPs were compared with hospitalization rates for residents of ZIP codes without such waste sites to look for evidence that potential exposures might be impacting health outcomes The studies report statistically significant elevations of cardiovascular disease hospitalizations in ZIP codes with POPs For coronary heart disease and a subset acute myocardial infarction the study showed elevations of approximately 15 and 20 respectively (Sergeev and Carpenter 2005) for stroke a 15 elevation (Shcherbatykh et al 2005) and for hypertension a 19 elevation (Huang et al 2006) in ZIP codes containing waste sites with POPs compared with hospitalization rates in other New York State ZIP codes (excluding NYC) that did not contain waste sites with POPs (Table 1)

In sub-analyses in these studies cardiovascular hospitalization rates for ZIP codes adjacent to the Hudson River were compared with cardiovascular hospitalization rates for ZIP codes in the rest of the state (excluding NYC) identified as not having inactive hazardous waste sites with POPs These sub-analyses showed statistically significantly elevated hospitalization rates of 36 and 39 for coronary heart disease and the subset acute myocardial infarction a 20 elevation for stroke and a 14 elevation for hypertension (Table 1) The Sergeev and Carpenter Shcherbatykh et al and Huang et al articles state that the Hudson River ZIP codes are in an area where people have higher income smoke less exercise more and have healthier diets This information is cited in these articles as evidence that known risk factors related to lower socioshy

1

economic status are likely not the cause for the elevated hospitalization rates in these Hudson River area ZIP codes

In these previously published reports Behavioral Risk Factor Surveillance System (BRFSS) data for entire counties are provided to support statements about the population of Hudson River ZIP codes having higher income and healthier lifestyles However because the ZIP codes adjacent to the Hudson River include only a proportion of the population of the counties abutting the river these county-level BRFSS data may be misleading regarding characteristics of the ZIP codes adjacent to the river (Figure 1) In fact data from the US Census show that the ZIP codes adjacent to the river have lower median income than the counties within which they are located The median income for the total population of the twelve counties adjacent to the Hudson River was $61010 in 2000 while the median income for the ZIP codes adjacent to the river was $45356 This median income for the population of the ZIP codes adjacent to the river is also slightly lower than the statewide (excluding NYC) median income of $47735 (US Census 2000) Thus the Hudson River ZIP codes analyzed in the Sergeev and Carpenter Shcherbatykh et al and Huang et al studies include a population of lower not higher socio-economic status compared to the rest of NYS (excluding NYC)

Because CVD and many CVD risk factors are associated with lower socio-economic status it is especially important to adjust as much as possible for socio-economic status when conducting comparative analyses of CVD hospitalizations or disease incidence or prevalence Regarding the hypothesis about POPs and CVD research studies described in more detail below have suggested possible biological mechanisms for POP exposuresrsquo effects on the heart There is very little research on human exposures to POPs and potential cardiovascular effects however

In response to Congressman Hincheyrsquos request for follow-up of the 2004-2005 findings ATSDR and NYS DOH agreed to conduct a similar study also ecological but at a different level of geography the Census block group By using a level of geography that is generally smaller in area and population than a ZIP code this follow-up seeks to more accurately adjust for socioshyeconomic and other factors that affect cardiovascular hospitalizations and to improve upon the geographic classification of relative distance to the Hudson River In addition rather than comparing hospitalizations among residents of Hudson River area ZIP codes to hospitalizations among residents of all other NYS (exclusive of NYC) ZIP codes with no waste sites with POPs this follow-up investigationrsquos analyses are restricted to the twelve counties abutting the Hudson river from Washington County south to Westchester County Within the study arearsquos twelve counties cardiovascular hospitalization rates in areas closest to the Hudson River are compared with rates in areas farther from the river This more localized approach may also assist with reducing the potential impact of regional differences in medical care practice or reporting factors that could affect the hospitalization data (Ko et al 2007) (Table 2 shows a more detailed listing of the methods used in the prior studies and follow-up study)

Regarding environmental exposures it is important to note that living near a hazardous waste site that contains PCBs or other POPs does not necessarily mean that human exposure from the site has occurred Depending upon the conditions at a site or the specific chemicals exposures may be highly improbable Information on body burden for example PCB levels in blood or on levels of the chemical in off-site air dust or soil is needed to support a claim of exposure The

2

studies published in 2005-2006 did not provide specific information such as environmental or biomonitoring data in support of the assumption that residence in ZIP codes containing hazardous waste sites with POPs or residence in ZIP codes adjacent to the Hudson River resulted in increased exposures to POPs in the population For the purpose of this follow-up we did not conduct environmental or human biological sampling to address this issue Rather this follow-up replicates the prior studiesrsquo approach with some enhancements to better control for confounding particularly confounding by socio-economic status

This descriptive epidemiologic investigation uses already existing hospitalization data from the Statewide Planning and Research Cooperative System and population data from the United States Census to evaluate comparative rates of cardiovascular hospitalizations While this type of investigation can not show proof of cause and effect it can assist in interpreting relationships among indicators of CVD burden and hypothesized exposures associated with distance from the Hudson River This descriptive investigation may also assist in determining whether additional health investigations of these associations are warranted

2 BACKGROUND

Cardiovascular Disease

The widely known risk factors for cardiovascular disease (CVD) in addition to age sex race and family history include tobacco use high blood pressure high serum cholesterol alcohol use obesity low fruit and vegetable intake physical inactivity diabetes low socio-economic status mental ill-health psychosocial stress and the use of certain medications (WHO 2010 NYS DOH 2009 Wing 1988 NYS DOH 2009a)

CVD Definitions CVD is a general term for any disease of the circulatory system The two main types of CVD are ischemic heart disease also termed coronary heart disease and cerebrovascular disease also called stroke Ischemic heart disease (IHD) results from restricted blood flow through the arteries supplying the heart muscle The most serious risk from IHD is acute myocardial infarction (AMI) another term for heart attack AMI occurs when blood flow to the heart is suddenly reduced or stopped usually due to a blockage in a coronary artery Angina is severe constricting pain in the chest often radiating from the area immediately over the heart to the left shoulder and down the arm Cerebrovascular disease refers primarily to stroke the sudden impairment of brain function resulting from interruption of circulation to the brain Ischemic stroke results from blockage of an artery supplying blood to the brain Hemorrhagic stroke occurs due to the escape of blood from a ruptured artery supplying blood to the brain

CVD Burden and Trends CVD is the leading cause of death for men and women of all ages in the United States and in New York State In 1999 for every death from IHD there were five hospitalizations for ischemic disease in NYS (NYS DOH 2009a) Mortality rates from CVD have been declining steadily since the 1980rsquos Research studies estimate that approximately half of the decline is due to improved treatments for CVD and its risk factors such as high cholesterol and high blood pressure and approximately half is due to decreasing incidence of CVD The decrease in incidence is due to trends in health behaviors primarily reduced smoking rates

3

CVD mortality rates do not appear to be declining for all segments of the population however and may actually be increasing among some groups particularly people of lower socio-economic status (Barnett et al 1999) This may be due in part to higher rates of smoking among people of lower socio-economic status as well as to other factors related to socio-economic status and race A growing literature points to multiple causes for CVD and other health disparities including health practices psychosocial stress limited resources discrimination and access to health care (Health United States 2007) Increases in the prevalence of diabetes and obesity in the general population partially offset advances in treatment and favorable trends for other risk factors An increasing ischemic heart disease mortality rate in women ages 35 to 54 since 1997 is attributed to these unfavorable risk factor trends (Ford 2007 Lloyd-Jones 2009) New York State (NYS) and other northeast states have high IHD and low stroke mortality with NYS having the highest IHD mortality and lowest stroke mortality among the 50 states (Howard et al 2009)

CVD Data Sources CVD mortality rates are more easily measured than incidence prevalence or hospitalizations because mortality data from vital records are a comprehensive and relatively accessible data source Information for estimating the incidence and prevalence of CVD outcomes comes primarily from health interview physical examination and record surveys Hospitalization data are available for NYS and the US but are generally collected and analyzed in order to track trends in hospital care and resources expended Actual CVD incidence data validated by physician diagnosis physician examination or reviews of medical records are not directly available for the entire US or for specific geographic areas of NYS Such information is available from population-based cohort studies that follow people over time in selected communities While the concern underlying this research project is to understand CVD levels in the community ie incidence and prevalence we do not have data that directly provide us with this information In the absence of such data hospitalizations are frequently used as indicators of the burden of disease in communities particularly in surveillance or hypothesis-generating investigations such as this one In NYS the Behavioral Risk Factor Surveillance System (BRFSS) an ongoing monthly telephone survey of adults aged 18 years and older provides information about the burden of CVD by age sex race income and education

CVD Risk Factor Data The BRFSS data for NYS show that over 20 of the population aged 65 and older reported having been diagnosed with CVD versus less than five percent of the population under 65 Males were slightly more likely to report having CVD than females Nonshyhispanic whites compared to other ethnic groups combined were slightly more likely to report having CVD None of these slight differences were statistically significant however The only statistically significant demographic risk factors in the BRFSS data were low income and education levels Statistically significant elevations of reported CVD were observed for respondents with incomes below $25000 and respondents with a high school degree or less education (NYS DOH 2009a)

Another factor that can affect the outcome measure hospitalizations is the likelihood of mortality prior to reaching the hospital Studies show a variety of factors affect whether an individual experiencing a heart attack or stroke outside the hospital survives to be counted in hospitalization data These factors include lack of knowledge about personal risk or prior disease socio-economic status race marital status and distance to the hospital (Galea 2007

4

Barnett 2006 Ayala 2003) Population-based studies report sudden cardiac death usually outside the hospital to be the first overt manifestation of heart disease for 40 to 60 of all cases ( Kannel 1987 deVreede-Swagemakers 1997 UK Heart Attack Study 1998) An analysis of out-of-hospital mortality from AMI in Pennsylvania in 1998 showed that in relatively rural areas where travel time to the hospital was 25 minutes or longer 72 of all AMI deaths occurred outside the hospital compared with a statewide average of 49 of AMI deaths occurring outside of the hospital (OrsquoNeill 2003) In New York State 55 of IHD deaths occur outside the hospital compared to 35 of stroke deaths (NYS DOH 2009b)

Hudson River PCB Contamination

During an approximate 30-year period ending in 1977 the General Electric (GE) plants in Hudson Falls and Fort Edward used PCBs in the manufacture of electrical capacitors and discharged large quantities of PCBs into the upper Hudson River Estimates of the total direct discharges to the river are as high as 1330000 pounds In 1983 the US EPA classified a 200shymile section of the Hudson River from Hudson Falls to New York City as a National Priority List site making it one of the largest Superfund sites in the United States In 2002 the US EPA issued a Record of Decision calling for targeted environmental dredging and removal of PCB-contaminated sediment in the upper River (USEPA 2002) The dredging began in May 2009

The Superfund environmental investigations and planned remedial actions have focused on the areas of the river near Hudson Falls and Fort Edward where the highest levels of PCBs in sediments are located The 2002 Record of Decision provides information about the human health risk assessment conducted for the Upper and Mid-Hudson and the ecological risk assessment conducted for the Upper and Lower Hudson River (USEPA 2002) Fish contamination and fish consumption are the primary concerns at the site

In 1976 fishing was banned in the upper river and in 1995 a catch and release policy replaced the ban NYS DOH continues to recommend people eat no fish from the Upper Hudson River that children under age 15 and women of childbearing age eat no fish from the entire 200 mile stretch of river below Hudson Falls and that the general population eat none of most species of fish caught between the federal dam at Troy and Catskill approximately 40 miles south of Troy The Lower Hudson remains closed to commercial fishing for striped bass and eight other species

The fishing bans and advisories were issued because ingestion of fish is the major potential pathway for exposure for the general population Many previous studies have shown relationships between consumption of fish from contaminated waters and human PCB levels Another potential pathway for exposure less frequently studied to date is inhalation of PCBs NYS DOH and ATSDR launched ldquoThe Hudson River Communities Projectrdquo in 2000 to address this issue (More information on this project is provided below)

POPS and PCBs

The category ldquopersistent organic pollutantsrdquo (POPs) includes man-made chemicals that are by definition persistent in the environment Examples of POPs are polychlorinated biphenyls

5

(PCBs) polychlorinated dibenzo-p-dioxins (PCDDs) polychorinated dibenso-furans (PCDFs) and pprsquo-diphenyldichloroethene (DDE) a breakdown product of the pesticide dichlorodiphenyltrichloroethane (DDT) These chemicals are stored in human fat tissues accumulating over an individualrsquos lifetime They also accumulate in fish and other species thereby concentrating up the food chain Levels of most POPs measured in blood serum of Americans have decreased in the last few decades An exception is polybrominated diphenyl ethers (PBDEs) levels of which have been increasing in recent years While levels of many POPs are declining a large number continue to be detectable in blood for most of the US population (USDHHS 2009)

This follow-up investigation focuses on the Hudson River where the POPs of concern are PCBs PCBs are a group of synthetic chlorinated compounds manufactured in the US through 1977 Production was halted due to concerns about the persistence of PCBs in the environment and the toxicity of PCBs shown in animal models PCBs were used primarily by the electrical industry for capacitors and transformers and were also used in hydraulic fluids fluorescent light fixtures flame retardants inks adhesives carbonless copy paper paints pesticide extenders plasticizers and wire insulators PCBs can be oily liquids or solids and some are volatile and can exist as vapor in air PCBs in the environment are generally mixtures of the many different types known as congeners (ATSDR 2000)

The variety of types of PCBs adds to the complexity of studying adverse animal and human health effects Increasingly PCB research focuses on specific PCB congeners and groups of congeners because specific types of congeners appear to have different mechanisms of action and toxicity Dioxin-like PCBs (DLPCBs) are an important subset of PCBs due to concerns about their toxicity The focus on dioxin-like PCBs is paralleled by increasing research on the group of POPs with dioxin-like properties This type of PCB and POP research assigns toxic potency values relative to dioxin to various PCBs and POPs to provide an overall toxic equivalency value to specific PCB and other POPs

From 2002 through 2007 NYS DOH and SUNY Albany School of Public Health researchers conducted the Hudson River Communities Study among residents of Hudson Falls and Fort Edward the villages where the PCB discharges from two GE electrical capacitor plants occurred The comparison community was upriver from the discharges in Glens Falls This studyrsquos evaluation of exposures included measuring PCBs in outdoor air and in blood serum The study found that PCB levels in outdoor air were statistically significantly higher in the Hudson FallsFort Edward area (072 nanograms per cubic meter) than in the upriver Glens Falls area (040 nanograms per cubic meter) but that the PCB concentrations in the Hudson FallsFort Edward area were lower than those in other communities with known PCB-contaminated sites and similar to levels reported in other locations in the northeastern United States A statistically significant reduction in the total PCB concentrations in outdoor air samples was observed at a distance of 1200 meters (34 mile) from the river from 0760 ngm3 to 0497 ngm3 (Fitzgerald et al 2007 Palmer et al 2008)

The average PCB concentrations in serum for the study and comparison populations did not differ significantly (307 ppb wet weight versus 323 ppb) The average toxic equivalency quantities (TEQ) for dioxin-like PCB congeners for the study and comparison area populations

6

also did not differ In addition the study found no detectable differences in study subjectsrsquo serum PCB levels associated with proximity to contaminated portions of the river or other PCB sources The study did find that lifetime consumption of Hudson River fish was associated with higher serum PCB levels (Fitzgerald 2007 Palmer 2008) Another recent biomonitoring study of Brooklyn Manhattan and New Jersey fisherman who consumed versus those who did not consume their catch from the lower Hudson River and New YorkNew Jersey harbor suggests eating fish from these waters is not associated with an increased body burden of PCBs or most other organochlorines This study did not address the inhalation pathway (Morland 2008)

PCBs and Cardiovascular Disease Outcomes

Research studies on adverse health effects associated with PCB exposure have focused on many types of outcomes organs and biological systems including cancer liver reproductive and developmental effects thyroid endocrine immune system and neurological effects as well as specific effects on the skin and eyes Relatively little research has focused to date on associations between PCB exposures and CVD outcomes The majority of human studies have assessed health effects in occupational groups having relatively high-level exposures among people consuming fish from contaminated waters or people exposed to relatively high levels due to accidental contamination of food (USDHHS NIH 2010)

PCBs have been shown to produce a variety of toxic effects in animals These effects include cancer allergies hypersensitivity damage to the central and peripheral nervous systems reproductive disorders and disruption of the immune system (ATSDR 2000) In contrast research to date does not tell us whether specific types of human health effects have occurred from the low-level exposures experienced by the general population Some human epidemiological studies of relatively low levels of PCDDs PCDFs and PCBs suggest possible relationships not only with neurodevelopment of infants but also diabetes and a condition called insulin resistance conditions that increase CVD risk (Longnecker 2001 Carpenter 2006 Arisawa 2005)

In 1998 a study of a population in Seveso Italy exposed to relatively pure TCDD (dioxin) from an industrial accident showed increased mortality from CVD respiratory diseases and diabetes (Bertazzi 1998) These findings suggested the possibility that other dioxin-like chemicals such as dioxin-like PCBs might increase risk for these diseases Some more recent animal studies described below suggest possible mechanisms of action for PCB effects that increase risk for CVD diabetes and hypertension Another relevant source of information is from research using the biological monitoring data from the National Health and Nutrition Examination Survey (NHANES) project to look at the association between PCB levels in blood and self-reported diabetes and heart disease The body of evidence based on analytical human epidemiological studies remains relatively sparse however

Some recent studies using NHANES biomonitoring data show associations between PCB levels and CVD (Lee 2006 Lee 2007 Ha 2007 Everett 2008) These studies are cross-sectional (comparisons at one point in time) evaluations of PCB levels measured in blood samples from individuals and self-reported CVD and diabetes a risk factor for CVD While these studies use individual-level biomonitoring and self-reported health outcome data they are considered to be

7

relatively weak in terms of providing evidence of causality because both exposure and outcome are measured at the same point in time Given the nature of NHANES data collection and reporting practices these studies are not able to control for possible geographical confounding which may occur if urban versus rural residence is associated with higher PCB levels and higher risk for CVD for example The higher PCB levels among African Americans in these data suggest urban residence may be playing a role in the PCB data

The results of animal studies suggest multiple mechanisms by which PCB exposures could affect cardiovascular risk as well as multiple types of adverse health outcomes One hypothesized mechanisms of action specifically for dioxin-like PCBs is that they may have anti-estrogenic effects which impact CVD risk factors such as serum cholesterol levels and blood pressure (Lind 2004) Several studies suggest PCBs interact with a specific receptor (AhR) and increase cellular oxidative stress causing inflammation cell injury and cardiovascular dysfunction (Hennig 2002) Similarly some studies suggest PCB-induced inflammation of fat cells leads to obesity and increased risk for coronary artery disease (atherosclerosis) (Arsenescu 2008)

3 FOLLOW-UP STUDY DATA SOURCES AND METHODS

This follow-up investigation will provide perspective on the 2004-2005 findings (Sergeev and Carpenter 2005 Shcherbatykh et al 2005) by producing results from analyses conducted at a more detailed level of geography the block group rather than ZIP code in order to better control for confounding When the term confounding is used by epidemioogists it refers to the problem that can arise when there is a factor such as income level that is not the factor of interest in the study but that is related to the risk factor being studied (residential location ndash proxy for exposure) and to the disease being studied (CVD) This type of factor is labeled a ldquoconfounderrdquo because it can confound ie interfere with or confuse the evaluation of associations among the variables of interest in the data All types of epidemiological studies face issues of confounding but ecological ie group-level analyses face additional challenges related to potential confounding because analyses can not account for some important individual differences within the groups that are the units of analysis (Morganstern 1998b Hertz-Picciotto 1998) This investigation was able to control for each individualrsquos age sex and race but not for individual-level risk factors such as income or education level

The follow-up investigationrsquos specific objectives are to

improve the ability to control for confounding by socio-economic status through the use of block groups rather than ZIP codes as the unit of analysis

compare the hospitalization rates among residents of block groups nearest the Hudson River to those among residents of block groups farther from the Hudson River

compare the findings of this follow-up study to those of the earlier studies

Study Area and Exposure Areas

The overall study area is comprised of the 12 counties with geographic areas abutting the Hudson River south of the source of PCB contamination from the General Electric facility in Hudson Falls The 12 counties are Westchester Rockland Putnam Orange Dutchess Ulster

8

Columbia Greene Rensselaer Albany Washington and Saratoga US Census boundaries are used to define the study area because census population estimates provide the data for the study area populationrsquos size and the populationrsquos age sex and race distribution Because the study period spanned 1990 through 2005 and populations change over time 1990 and 2000 Census data were used to linearly interpolate the between-year populations from 1991 through 1999 for each block group 2000 data were used without extrapolation to estimate block group populations from 2001 through 2005

Because block group boundaries change between the Census years 1990 block populations were normalized to fit 2000 block group boundaries before the estimation of sex age and race group specific populations In the digital map 2000 block group boundaries were overlaid on the 1990 block centroids to assign 1990 block population counts to the corresponding 2000 block group boundaries Then 1990 block populations were summed to fit the block group boundaries Next population counts by age group (0-14 15-24 25-34 35-44 45-54 55-64 65-74 75+) sex (male female) and race (white black asian other) were estimated for each year at the block group level This investigation includes all ages over 25 years and the study population is divided into four race categories white black or African-American Asian and ldquoOtherrdquo The Pacific Island group was combined with the Asian group and American Indian ldquootherrdquo and ldquomultiple racesrdquo categories are combined into a larger ldquootherrdquo category

Classification of the population into potential exposure categories for this study is based exclusively on location relative to the Hudson River Two types of exposure proxy categories were developed block group adjacency and block group centroid distance to the river shore For the adjacency categories block groups were assigned to the adjacent category if they directly abutted the Hudson River and all other block groups in the 12 counties were assigned to the nonshyadjacent category

For the distance categories a central point (centroid) within each block group was identified and used to assign block group centroid distance to the river shore The block group centroids were population-weighted which means the geographic distribution of households within 2000 census blocks were used to locate block group centroids based on the geographic population distribution By locating a population-weighted block centroid rather than a more straightforward distance centroid the assignment of block group distance to the river shore more accurately reflects distance to the river for the majority of block group residents Buffers of one-half mile and one mile from the Hudson River boundary were created for the assignment of each block group into one of the three distance categories less than frac12 mile (closest) frac12 to one mile (close) and farther than one mile from the river (distant)

Distance categories were selected a priori at the start of the study Because environmental sampling data were not collected along the length of the river included in this investigation we are not able to determine whether these distance categories represent actual differences in exposures As discussed previously a statistically significant reduction in the total PCB concentrations in outdoor air samples was observed at a distance of 1200 meters (34 mile) from the river from 0760 ngm3 to 0497 ngm3 in a study conducted in the Hudson Falls Fort Edward area near the source of PCBs (Fitzgerald 2007 Palmer 2007)

9

For the adjacency categories of the 2371 block groups in the study area 255 (11) were classified as adjacent to the Hudson River and 2116 (89) were classified as non-adjacent (Figure 2) For the distance categories 225 (9) were classified as closest to the river 216 (9) as close and 1930 block groups (81) as distant from the river

Comparison of Figures 1 and 2 show that block groups include smaller geographic areas than ZIP codes and that block groups are relatively smaller in area in urban versus rural areas The study area includes a total of 368 ZIP codes compared to a total of 2371 block groups ZIP code average population in the study area is approximately 8000 compared to block group average population of 1200 Comparison of Figures 2 and 3 shows that using population-weighted centroid distance to the river produces different exposure areas than adjacency Fifty-three (20) of the adjacent block groups were categorized for the distance indicator as distant because the population in the block group was concentrated in the area farther than one mile from the river On the other hand about 11 of the 2116 non-adjacent block groups were classified as closest (77) or close (162)

Hospitalization Data

Hospitalizations for ischemic heart disease (IHD) and cerebrovascular disease (stroke) were included in this study More specifically hospital discharge records that listed as the principal diagnosis either IHD (ICD-9-CM 410-414 and 4297) or cerebrovascular disease (ICD-9-CM 430-436) and subsets of each of these categories similar to the subsets evaluated in the Sergeev et al and Shcherbatykh studies were included The subsets are AMI and angina subsets of IHD and ischemic and hemorrhagic stroke which are subsets of cerebrovascular disease (stroke) (Principal diagnosis represents the illness for which the person was admitted to the hospital) Angina is an outcome not evaluated in the prior studies It was added here for comparative purposes because it represents an outcome appropriately addressed by primary outpatient rather than inpatient hospital care Angina is one of a group of outcomes included in the recently developed national ldquoPrevention Quality Indicatorsrdquo (USDHHS AHRQ 2007) as an indicator for geographic areas or populations for whom primary care is inadequate Table 3 lists the specific ICD-9-CM codes and definitions included in this investigation

While one of the previously published studies (Huang et al 2006) showed elevations in hypertension (high blood pressure) rates this outcome is rarely listed as the principal reason for hospitalization As a secondary diagnosis however hypertension is extremely common with substantial overlap with the primary diagnoses of IHD and stroke In addition hypertension hospitalizations were not as elevated in the ZIP codes adjacent to the Hudson River as the IHD and stroke outcomes As a practical matter geocoding the hypertension hospitalization data (1364000 hospitalizations) would overwhelm the resources for this project For these reasons hypertension was not included in this follow-up investigation

The source of the hospitalization data was the NYS DOH Statewide Planning and Research Cooperative System (SPARCS) established in 1979 to collect detailed records on discharges from hospitals located in New York State Persons seen in the emergency department but not admitted are not included in the SPARCS data Hospitalizations at US Veterans Administration

10

Hospitals are also not included in this dataset We identified the hospital transfers in the data and excluded them so that individuals are only counted once for the same event

The NYS DOH Data Protection Review Board for SPARCS hospitalization data approved our use of hospitalization records for this investigation and provided approximately 460000 records with a principal diagnosis of either IHD or stroke Because of the large number of cardiovascular hospitalizations and this studyrsquos requirement for geocoding each hospitalized personrsquos address we selected a 50 random sample of the hospital discharges with principal diagnoses of IHD or stroke for inclusion in the study (229948 records) Approximately one percent of these records (2737) were eliminated from the study because they contained no residential address resulting in 227211 hospitalization records 24939 of these records were eliminated after geocoding because they were not within the 12 county study area This left 202272 records remaining in the study

For geocoding to the block group level the hospitalization street addresses were modified as needed using US Postal Service standards The addresses were assigned geographic coordinates using commercially available geocoding software (MapMarker 2004) A combination of land parcel data and street centerline files were used to assign geographic coordinates No contact was made with cases parents legal guardians or next of kin of cases to determine residential locations

Once geographic coordinates were assigned to cases through address-matching the case locations were overlaid onto digital maps of the study area using a geographic information system so that the number of observed cases falling within the study area boundaries could be determined and a 2000 Census block group could be assigned to each hospitalization In order to protect confidentiality no maps of individual case locations are provided Geographic coordinates were not able to be assigned for some addresses for example rural routes or post office boxes Internet searches were conducted for addresses that were located in facilities such as nursing homes and trailer parks

While fewer than one percent of records were missing sex or age information approximately 15000 records seven percent were missing race information Evaluation of the geographic and temporal distribution of these records showed they were not randomly geographically distributed 47 of the records with no race information were from one hospital for that hospital race was missing from 100 of the records in all years except one after 1997 mid-way through the study time-frame Evaluation of other records missing race showed similar patterns of 100 of records missing race in specific years In order to prevent biased hospitalization rate estimates records missing race were assigned race randomly assuming a race distribution equal to that of the individualrsquos residential block group

Statistical Analysis

This investigation used negative binomial multivariable regression models the statistical methods used in the previous studies (Sergeev and Carpenter 2005 [IHD] Shcherbatykh et al 2005 [stroke]) The follow-up studyrsquos analyses compared hospitalization rates for populations in block groups adjacent or in close proximity to the Hudson River to rates for populations more

11

distant from the Hudson River while adjusting for the potential influence of other factors Sex age and race-specific hospitalization count data comprise the dependent variable and the statistical methods attempt to predict the counts using the sex age and race distributions from Census data as well as block group-level demographics such as median income Models were first estimated using basic Census information for age sex and race Then more fully adjusted models were estimated using information at the block group level for income education population density Hispanic ethnicity and distance to the nearest hospital

More specifically each hospitalized personrsquos block group was used to assign quartile categories of block group median household income of population with less than a college education population density of the population identified as Hispanic and distance from the block group centroid to the nearest hospital These variables were selected in advance The population density variable was selected to account for potential effects on hospitalization rates of large differences in population density from the relatively rural northern counties to the more urban southern counties in the study area of the population identified as Hispanic was included because Hispanic ethnicity may be associated with CVD risk but Hispanic ethnicity was not available in the hospitalization records Distance to the hospital was included because studies show that shorter distances increase the likelihood that individuals with AMI and stroke survive to be admitted to the hospital

As described previously the literature about risk factors for CVD shows strong associations between lower income and education and higher CVD risk Regarding the effects of income and education on hospitalization rates the literature from analytic studies specifically of hospitalizations for CVD is quite sparse Most analytic studies focus on prevalence incidence or mortality in population-based studies or cohort studies and there are not many of these An example of a study that was useful as background is a study of income level and CVD mortality that showed strong associations of lower income and higher CVD mortality for men and women with the relationship for men being linear throughout the range of income levels while for women the relationship was only strong at levels of income below the median (Rehkopf 2008)

This follow-up studyrsquos approach was to use the same set of variables for the more fully adjusted models using quartile cut points for all the sexrace groups and types of CVD diagnostic subsets The 2000 Census data were used to create the population-weighted quartile categories to be used in the statistical analyses The quartile cut points are shown below

population density(persons per square mile)

median 1999 household income

with less than a college education

with Hispanic ethnicity

Hospital distance

Quartile 1 61 - lt507 $2499-lt$40968 0 - lt31 0 - lt2 lt25 (median) Quartile 2 507 - lt102 $40968 - lt54650 31 - lt43 2 - lt5 25 - lt6 (50-75) Quartile 3 102 - lt6410 $54650 - lt72944 43 - lt55 5 - lt10 6 ndash 11 (75-90) Quartile 4 6410 - 94372 $72944 - 200001 55 - 100 10 - 86 11 ndash 45 (90-100)

SAS version 9 provided the statistical programs used to estimate hospitalization rates for exposure proxy categories and demographic categories The models provide ldquocontrast

12

estimatesrdquo ie rate ratios for one category versus a reference category of a variable and 95 confidence intervals (CI) for these rate ratios For the location categories rate ratios for the areas adjacent are compared to areas not adjacent to the river (reference) and areas closest and close to the river are compared to areas distant from the river (reference) If a rate ratio is greater than 100 the model has estimated an elevation of hospitalizations in that category of the study population compared to the reference population If the rate ratio is less than 100 then there are fewer than expected hospitalizations compared to the reference population

The magnitude of the elevation or deficit is also estimated by the rate ratio For instance if the hospitalization rate is twice as high in the adjacent population it would result in a rate ratio of 200 while a 50 higher rate would result in a rate ratio of 150 If the hospitalization rate in the comparison group was only half as big as in the reference group this would result in a rate ratio of 050 The 95 confidence interval assists with determining whether the observed differences are statistically meaningful ie statistically significant The confidence interval provides the range in which there is a 95 probability of including the true rate ratio

In addition to directly assess potential confounding we examined the associations between income and exposure proxy categories and between income and CVD hospitalization risk Because the modeling process provides rate ratios that provide comparisons among groups but not the rates themselves we also calculated the rates per population for each sex race and income category in order to directly observe the relative magnitudes of CVD hospitalization rates Finally analyses stratified by income quartile were also conducted in order to assess the potential role of confounding by income The stratified analyses consist of negative binomial regressions conducted separately for each quartile of income

4 RESULTS

Demographics Associations with Proximity to the River

Study Population Table 4 provides detailed demographic characteristics for the total study population and the populations of the adjacent versus non-adjacent and nearer versus farther from the river block groups from the 2000 Census The total population of the study area for this one year of the study time period is 2924631 persons The first row of Table 4 shows that approximately 11 percent of the study area population is classified as living in block groups adjacent to the Hudson River and 89 live in non-adjacent block groups Approximately 8 of the studyrsquos population lives in the closest block groups approximately 10 lives in close block groups and approximately 82 in distant block groups The table also shows that approximately 81 of the total study population is identified as white 9 black 3 Asian less than 01 Pacific Islander 02 American Indian 4 ldquootherrdquo and 2 ldquomultiple racesrdquo

The 2000 Census data (Table 4) also show that the populations in block groups adjacent and nonshyadjacent to the river are similar in terms of the percent of minority population but the adjacent block groups include a smaller percentage of persons who have attended college and a higher percentage of households below the poverty level Median income is $47320 for block groups adjacent versus $62668 for block groups not adjacent to the river Data for the study area from the 1990 Census show similar distributions (data not shown)

13

For the distance to the river categories there is a gradient for the three levels of distance from the river with the closest block groups on average having higher percentages of minorities smaller percentages of persons who have attended college and higher percentages of households below the poverty level (Table 4) The populations of the close and closest block groups are more similar to each other than to the populations of the distant block groups The population of the closest block groups is estimated to be 19 black close block groups 14 black and distant block groups 8 black The proportions for the ldquootherrdquo category used in the analyses which includes multiple races and American Indians are approximately 12 in both the closest block groups and in the close block groups compared to 5 in the block groups most distant from the river Median income for the closest block groups is $40748 close block groups $44962 and most distant block groups $64847

The bar chart below illustrates the correlations between US Census 2000 indicators of block group socio-economic status and block group centroid (population weighted) distance from the river for the study population

2000 Census Demographics for Study Population by Distance to River Category

black other hispanic no college below poverty

of pop ulati on

0

10

20

30

40

50

60

lt12 mile

12 to 1 mile

gt1 mile

Table 5 shows selected demographic information for the person-year estimates used in the study analyses for the 1990-2005 study period Person-year data represent individuals from Census estimates counted repeatedly for each year of the study timeframe As described in the Methods section the estimation of person years used interpolation to account for population changes from 1990-2000 and techniques for taking account of block boundary changes The person-years estimates for demographic categories (complete data not shown) show similar patterns as the 1990 Census data (data not shown) and 2000 Census data (Table 5) Table 5 for example shows nearly identical percentages of population in each of the river distance categories as Table 4

Table 5 also shows person year information for income quartiles and exposure proxy categories as well as race and exposure proxy categories that is consistent with the Table 4 data for the

14

2000 population Table 5 row percentages show for example that while approximately 11 of the total study population live in adjacent block groups approximately 19 of the population in the lowest income quartile live in adjacent block groups By design approximately 25 of the total study population is in each of the income quartiles Table 5 column percentages show however that in the adjacent block groups 41 of the population is in the lowest income quartile and only 9 are in the highest income quartile

For the block groups within frac12 mile of the river the distribution among income quartiles is even more skewed towards the lowest income quartile 61 of the population living within frac12 mile of the river is in the lowest income quartile and only 6 is in the highest income quartile Given that approximately 25 of the total study population is within each income quartile these percentages indicate that living within the block groups closest to the river is associated with more than a doubling of the likelihood of being in a block group in the lowest socio-economic status quartile 436 of those living in the block groups close to the river (12 to one mile from the river) are in block groups in the lowest income quartile While less than a doubling this indicates that living in the close block groups is associated with a 75 increase in the likelihood of being in the lowest quartile for socio-economic status

Hospitalizations Table 6 shows the process of removing hospitalization records from the study for a variety of reasons The first column includes the 202272 records that contained a residential address and were located within the 12-county study area as described in the Methods section The next five columns show the records remaining after data cleaning exclusion of transfers elimination of records with missing covariates and exclusions due to poor geocoding results Approximately 33000 records were eliminated because they were transfers from one hospital to another for the same event For approximately 1100 records geocoding was inadequate for assigning block group for imputing race as described in the Methods section 336 records were missing sex (11) or age (325) Of the 167733 records remaining 154220 (92) were able to be adequately geocoded to a street address for assignment in a specific block group While approximately eight percent of hospitalizations were eliminated due to inadequate geocoding Table 6 shows that the percentages of hospitalization records for white black Asian and ldquootherrdquo categories and the distribution among diagnostic categories change very little after these records were dropped

The final column of Table 6 shows that of the 154220 hospitalizations in the analysis 101012 IHD hospitalizations comprise the largest health outcome category and are 65 of the hospitalizations in the study AMI hospitalizations (35811) account for 35 of the IHD hospitalizations and angina (4731) accounts for five percent of the IHD hospitalizations Stroke hospitalizations (53208) comprise 34 of the studyrsquos hospitalizations Approximately 36 of the stroke hospitalizations are for ischemic stroke (19371) and approximately 11 are for hemorrhagic stroke (5995)

Table 7 shows the distributions of hospitalizations among the four income quartiles While each income quartile (by definition) contains approximately 25 of the study population for all types of hospitalizations we see that the lowest income quartile includes more than 25 of hospitalizations from a minimum of 286 (IHD) to a maximum of 356 of hospitalizations (angina) The highest income quartile includes fewer than 25 of hospitalizations for each CVD

15

category ranging from 182 for angina to 224 for hemorrhagic stroke These deviations from the 25 quartile distribution show a strong linear pattern with the largest percentage of hospitalizations coming from residents of block groups in the lowest income quartile and the smallest percentages coming from residents of the most wealthy block groups For AMI the lowest income quartilersquos share of hospitalizations is 50 higher than the highest income quartilersquos share of hospitalizations (30 vs 20) IHD and stroke show similar differences and angina shows a doubling with the lowest income quartile contributing 36 of hospitalizations compared to the highest income quartilersquos 18 share of hospitalizations

Table 7 also shows the distribution of hospitalizations among the race categories and the study population race category distribution for comparison Blacks show lower percentages of IHD and AMI than their population percentages and higher percentages of angina and stroke While full presentation of race and sex findings is beyond the scope of this report the lower than expected percentages of IHD and AMI among blacks who on average have lower socioshyeconomic status and are at greater risk for CVD hospitalizations raised questions that were addressed by calculating hospitalization rates per 1000 population for each sex and race group for each income quartile (Appendix B Table 2) Figures 4 5 and 6 show these rates for the white black and Asian categories (These rates for the ldquootherrdquo category are comparatively very high and are not shown on these graphs This issue is addressed in the Discussion section) Figures 4 5 and 6 show the consistent and strong effects of income quartile on hospitalization rates for AMI and stroke for whites blacks and Asians of both sexes The other outcomes show similar patterns (The ldquootherrdquo group showed an opposite pattern [See Discussion Section])

Demographics Summary

The strong associations of lower income with residence closer to the river (Table 5) and lower income with higher hospitalization rates (Tables 7 amp 8) create a difficult challenge for this investigation The multivariable analyses of hospitalization rates attempt to control for socioshyeconomic differences However because these income differences which strongly affect hospitalization rates are also strongly correlated with the gradations of exposure (river distance) it will be difficult to disentangle the influence of socio-economic status from the influence of potential exposures

Statistical Analyses Proximity to the River and CVD Hospitalization Risk

All the variables described above were included in the multivariable regression models and almost every quartile category for every variable showed statistically significant and expected types of effects for all the CVD hospitalization categories (Appendix B Tables 1a and 1b) The following presentation of results focuses on the rate ratios for the exposure proxy variables Table 8a (IHD) and 8b (stroke) show the regression modeling results Results are shown first for models that include only age sex and race in addition to the exposure proxy categories and second for models that also include median income education population density Hispanic ethnicity and distance to nearest hospital The tables provide numbers of hospitalizations and person-years to assist with interpretation For example for total IHD we can see that while 108 of the study population lives in adjacent block groups 122 of the hospitalizations are among people in adjacent block groups This difference of 14 percentage points is the basis for

16

the finding of a statistically significantly elevated rate ratio (risk) for IHD among people living in the adjacent block groups (RR=133 CI 112-157)

Looking across the columns of Tables 8a and 8b we can see the rate ratios for each of the outcomes for the adjacent versus non-adjacent block groups and the results are similar to those for IHD The elevated rate ratios range from 123 for angina to 136 for AMI The only rate ratio not elevated in these first models adjusting only for age sex and race is for hemorrhagic stroke Going down the rows of the Table the next set of rate ratios are for the distance to the river categories The findings are very similar to those for the adjacent block groups Here however we have two exposure proxy categories allowing comparison of the two These rate ratios show no evidence that living in the block groups closest to the river (lt12 mile to the river) produces greater risk than living in the block groups close to the river (12 to 1 mile from the river)

The next set of rate ratios are for models with additional adjustment for the demographics of each hospitalized personrsquos block group as well as distance to the nearest hospital As expected adjustment for block group median income and other demographics accounts for some of the variability in hospitalization counts and thereby reduces the magnitude of the rate ratios estimated for the river adjacency and distance categories The rate ratio elevation for total IHD for example is reduced from 133 to 107 (RR=107 CI 104-111) The elevated rate ratios in these more fully adjusted models for the adjacency analyses are of similarly small magnitude ranging from 106 for AMI to 110 for ischemic stroke Four of the six outcomes show statistically significant elevations Again only hemorrhagic stroke shows no evidence of an elevated rate ratio

For the river distance categories none of the rate ratios show any evidence that living in the block groups closest to the river (lt12 mile to the river) produces greater risk than living in the block groups close to the river (12 to 1 mile) Of the six types of outcomes only two show statistically significantly elevated rate ratios for living in the closest block groups (total stroke and ischemic stroke) For the close block groups four out of six outcomes show statistically significantly elevated rate ratios The statistically significantly elevated rate ratios vary from 106 (AMI stroke) to 113 (angina)

To address the strong evidence described earlier for confounding in these data ie the association of lower income with both CVD hospitalizations and residence closer to the river we also conducted the analyses separately for each quartile of income These results are presented in Table 9 The first results are from analyses adjusting only for age sex and race and the second set of results are from models including the additional demographic factors For simplicity rate ratios are shown only for block groups closest to the river compared as in the other models to block groups farther than one mile from the river (Results were similar for the other distance categories (data not shown)

For the two lowest income quartiles Table 9 shows results similar to those for the study population as a whole (Table 8a 8b) However for the two highest income quartiles which represent frac12 of the study population Table 9 shows a consistent pattern of reduced rate ratios for residence in the closest block groups within frac12 mile of the river In the highest income quartile

17

1 INTRODUCTION

The New York State Department of Health (NYS DOH) has a cooperative agreement with the Agency for Toxic Substances and Disease Registry (ATSDR) to perform environmental exposure and health assessments conduct health statistics reviews and perform epidemiological studies of populations in New York State which may have been exposed to environmental contaminants In October 2006 Member of Congress Maurice Hinchey requested that ATSDR conduct a health consultation to assess cardiovascular disease among people residing near the Hudson River A 200-mile segment of the Hudson River from Hudson Falls south to New York City is a National Priority List (NPL) (Superfund) site due to contamination of the river with polychlorinated biphenyls (PCBs) from GE sites in Hudson Falls and Ft Edward (See Appendix A for Member of Congress Hincheyrsquos letter)

In support of his request Member of Congress Hinchey cited three studies (Sergeev and Carpenter 2005 [ischemic heart disease] Shcherbatykh et al 2005 [stroke] Huang et al 2006 [hypertension]) These statewide studies suggest residents of ZIP codes containing hazardous waste sites with persistent organic pollutants (POPs) such as PCBs have elevated risk for hospitalizations for cardiovascular disease including ischemic heart disease stroke and hypertension Analyses conducted separately for ZIP codes next to the Hudson River suggest residents of these specific ZIP codes also have increased rates of hospitalization for cardiovascular disease

These studies published in 2005-2006 are ecological studies In an ecological study data are analyzed by evaluating risk factors and health outcomes among groups For example rates of disease are compared for populations of ZIP codes with varying median income levels to see if income levels appear to be related to disease outcome rates In these previously conducted studies rates of hospitalization for residents of ZIP codes with hazardous waste sites containing POPs were compared with hospitalization rates for residents of ZIP codes without such waste sites to look for evidence that potential exposures might be impacting health outcomes The studies report statistically significant elevations of cardiovascular disease hospitalizations in ZIP codes with POPs For coronary heart disease and a subset acute myocardial infarction the study showed elevations of approximately 15 and 20 respectively (Sergeev and Carpenter 2005) for stroke a 15 elevation (Shcherbatykh et al 2005) and for hypertension a 19 elevation (Huang et al 2006) in ZIP codes containing waste sites with POPs compared with hospitalization rates in other New York State ZIP codes (excluding NYC) that did not contain waste sites with POPs (Table 1)

In sub-analyses in these studies cardiovascular hospitalization rates for ZIP codes adjacent to the Hudson River were compared with cardiovascular hospitalization rates for ZIP codes in the rest of the state (excluding NYC) identified as not having inactive hazardous waste sites with POPs These sub-analyses showed statistically significantly elevated hospitalization rates of 36 and 39 for coronary heart disease and the subset acute myocardial infarction a 20 elevation for stroke and a 14 elevation for hypertension (Table 1) The Sergeev and Carpenter Shcherbatykh et al and Huang et al articles state that the Hudson River ZIP codes are in an area where people have higher income smoke less exercise more and have healthier diets This information is cited in these articles as evidence that known risk factors related to lower socioshy

1

economic status are likely not the cause for the elevated hospitalization rates in these Hudson River area ZIP codes

In these previously published reports Behavioral Risk Factor Surveillance System (BRFSS) data for entire counties are provided to support statements about the population of Hudson River ZIP codes having higher income and healthier lifestyles However because the ZIP codes adjacent to the Hudson River include only a proportion of the population of the counties abutting the river these county-level BRFSS data may be misleading regarding characteristics of the ZIP codes adjacent to the river (Figure 1) In fact data from the US Census show that the ZIP codes adjacent to the river have lower median income than the counties within which they are located The median income for the total population of the twelve counties adjacent to the Hudson River was $61010 in 2000 while the median income for the ZIP codes adjacent to the river was $45356 This median income for the population of the ZIP codes adjacent to the river is also slightly lower than the statewide (excluding NYC) median income of $47735 (US Census 2000) Thus the Hudson River ZIP codes analyzed in the Sergeev and Carpenter Shcherbatykh et al and Huang et al studies include a population of lower not higher socio-economic status compared to the rest of NYS (excluding NYC)

Because CVD and many CVD risk factors are associated with lower socio-economic status it is especially important to adjust as much as possible for socio-economic status when conducting comparative analyses of CVD hospitalizations or disease incidence or prevalence Regarding the hypothesis about POPs and CVD research studies described in more detail below have suggested possible biological mechanisms for POP exposuresrsquo effects on the heart There is very little research on human exposures to POPs and potential cardiovascular effects however

In response to Congressman Hincheyrsquos request for follow-up of the 2004-2005 findings ATSDR and NYS DOH agreed to conduct a similar study also ecological but at a different level of geography the Census block group By using a level of geography that is generally smaller in area and population than a ZIP code this follow-up seeks to more accurately adjust for socioshyeconomic and other factors that affect cardiovascular hospitalizations and to improve upon the geographic classification of relative distance to the Hudson River In addition rather than comparing hospitalizations among residents of Hudson River area ZIP codes to hospitalizations among residents of all other NYS (exclusive of NYC) ZIP codes with no waste sites with POPs this follow-up investigationrsquos analyses are restricted to the twelve counties abutting the Hudson river from Washington County south to Westchester County Within the study arearsquos twelve counties cardiovascular hospitalization rates in areas closest to the Hudson River are compared with rates in areas farther from the river This more localized approach may also assist with reducing the potential impact of regional differences in medical care practice or reporting factors that could affect the hospitalization data (Ko et al 2007) (Table 2 shows a more detailed listing of the methods used in the prior studies and follow-up study)

Regarding environmental exposures it is important to note that living near a hazardous waste site that contains PCBs or other POPs does not necessarily mean that human exposure from the site has occurred Depending upon the conditions at a site or the specific chemicals exposures may be highly improbable Information on body burden for example PCB levels in blood or on levels of the chemical in off-site air dust or soil is needed to support a claim of exposure The

2

studies published in 2005-2006 did not provide specific information such as environmental or biomonitoring data in support of the assumption that residence in ZIP codes containing hazardous waste sites with POPs or residence in ZIP codes adjacent to the Hudson River resulted in increased exposures to POPs in the population For the purpose of this follow-up we did not conduct environmental or human biological sampling to address this issue Rather this follow-up replicates the prior studiesrsquo approach with some enhancements to better control for confounding particularly confounding by socio-economic status

This descriptive epidemiologic investigation uses already existing hospitalization data from the Statewide Planning and Research Cooperative System and population data from the United States Census to evaluate comparative rates of cardiovascular hospitalizations While this type of investigation can not show proof of cause and effect it can assist in interpreting relationships among indicators of CVD burden and hypothesized exposures associated with distance from the Hudson River This descriptive investigation may also assist in determining whether additional health investigations of these associations are warranted

2 BACKGROUND

Cardiovascular Disease

The widely known risk factors for cardiovascular disease (CVD) in addition to age sex race and family history include tobacco use high blood pressure high serum cholesterol alcohol use obesity low fruit and vegetable intake physical inactivity diabetes low socio-economic status mental ill-health psychosocial stress and the use of certain medications (WHO 2010 NYS DOH 2009 Wing 1988 NYS DOH 2009a)

CVD Definitions CVD is a general term for any disease of the circulatory system The two main types of CVD are ischemic heart disease also termed coronary heart disease and cerebrovascular disease also called stroke Ischemic heart disease (IHD) results from restricted blood flow through the arteries supplying the heart muscle The most serious risk from IHD is acute myocardial infarction (AMI) another term for heart attack AMI occurs when blood flow to the heart is suddenly reduced or stopped usually due to a blockage in a coronary artery Angina is severe constricting pain in the chest often radiating from the area immediately over the heart to the left shoulder and down the arm Cerebrovascular disease refers primarily to stroke the sudden impairment of brain function resulting from interruption of circulation to the brain Ischemic stroke results from blockage of an artery supplying blood to the brain Hemorrhagic stroke occurs due to the escape of blood from a ruptured artery supplying blood to the brain

CVD Burden and Trends CVD is the leading cause of death for men and women of all ages in the United States and in New York State In 1999 for every death from IHD there were five hospitalizations for ischemic disease in NYS (NYS DOH 2009a) Mortality rates from CVD have been declining steadily since the 1980rsquos Research studies estimate that approximately half of the decline is due to improved treatments for CVD and its risk factors such as high cholesterol and high blood pressure and approximately half is due to decreasing incidence of CVD The decrease in incidence is due to trends in health behaviors primarily reduced smoking rates

3

CVD mortality rates do not appear to be declining for all segments of the population however and may actually be increasing among some groups particularly people of lower socio-economic status (Barnett et al 1999) This may be due in part to higher rates of smoking among people of lower socio-economic status as well as to other factors related to socio-economic status and race A growing literature points to multiple causes for CVD and other health disparities including health practices psychosocial stress limited resources discrimination and access to health care (Health United States 2007) Increases in the prevalence of diabetes and obesity in the general population partially offset advances in treatment and favorable trends for other risk factors An increasing ischemic heart disease mortality rate in women ages 35 to 54 since 1997 is attributed to these unfavorable risk factor trends (Ford 2007 Lloyd-Jones 2009) New York State (NYS) and other northeast states have high IHD and low stroke mortality with NYS having the highest IHD mortality and lowest stroke mortality among the 50 states (Howard et al 2009)

CVD Data Sources CVD mortality rates are more easily measured than incidence prevalence or hospitalizations because mortality data from vital records are a comprehensive and relatively accessible data source Information for estimating the incidence and prevalence of CVD outcomes comes primarily from health interview physical examination and record surveys Hospitalization data are available for NYS and the US but are generally collected and analyzed in order to track trends in hospital care and resources expended Actual CVD incidence data validated by physician diagnosis physician examination or reviews of medical records are not directly available for the entire US or for specific geographic areas of NYS Such information is available from population-based cohort studies that follow people over time in selected communities While the concern underlying this research project is to understand CVD levels in the community ie incidence and prevalence we do not have data that directly provide us with this information In the absence of such data hospitalizations are frequently used as indicators of the burden of disease in communities particularly in surveillance or hypothesis-generating investigations such as this one In NYS the Behavioral Risk Factor Surveillance System (BRFSS) an ongoing monthly telephone survey of adults aged 18 years and older provides information about the burden of CVD by age sex race income and education

CVD Risk Factor Data The BRFSS data for NYS show that over 20 of the population aged 65 and older reported having been diagnosed with CVD versus less than five percent of the population under 65 Males were slightly more likely to report having CVD than females Nonshyhispanic whites compared to other ethnic groups combined were slightly more likely to report having CVD None of these slight differences were statistically significant however The only statistically significant demographic risk factors in the BRFSS data were low income and education levels Statistically significant elevations of reported CVD were observed for respondents with incomes below $25000 and respondents with a high school degree or less education (NYS DOH 2009a)

Another factor that can affect the outcome measure hospitalizations is the likelihood of mortality prior to reaching the hospital Studies show a variety of factors affect whether an individual experiencing a heart attack or stroke outside the hospital survives to be counted in hospitalization data These factors include lack of knowledge about personal risk or prior disease socio-economic status race marital status and distance to the hospital (Galea 2007

4

Barnett 2006 Ayala 2003) Population-based studies report sudden cardiac death usually outside the hospital to be the first overt manifestation of heart disease for 40 to 60 of all cases ( Kannel 1987 deVreede-Swagemakers 1997 UK Heart Attack Study 1998) An analysis of out-of-hospital mortality from AMI in Pennsylvania in 1998 showed that in relatively rural areas where travel time to the hospital was 25 minutes or longer 72 of all AMI deaths occurred outside the hospital compared with a statewide average of 49 of AMI deaths occurring outside of the hospital (OrsquoNeill 2003) In New York State 55 of IHD deaths occur outside the hospital compared to 35 of stroke deaths (NYS DOH 2009b)

Hudson River PCB Contamination

During an approximate 30-year period ending in 1977 the General Electric (GE) plants in Hudson Falls and Fort Edward used PCBs in the manufacture of electrical capacitors and discharged large quantities of PCBs into the upper Hudson River Estimates of the total direct discharges to the river are as high as 1330000 pounds In 1983 the US EPA classified a 200shymile section of the Hudson River from Hudson Falls to New York City as a National Priority List site making it one of the largest Superfund sites in the United States In 2002 the US EPA issued a Record of Decision calling for targeted environmental dredging and removal of PCB-contaminated sediment in the upper River (USEPA 2002) The dredging began in May 2009

The Superfund environmental investigations and planned remedial actions have focused on the areas of the river near Hudson Falls and Fort Edward where the highest levels of PCBs in sediments are located The 2002 Record of Decision provides information about the human health risk assessment conducted for the Upper and Mid-Hudson and the ecological risk assessment conducted for the Upper and Lower Hudson River (USEPA 2002) Fish contamination and fish consumption are the primary concerns at the site

In 1976 fishing was banned in the upper river and in 1995 a catch and release policy replaced the ban NYS DOH continues to recommend people eat no fish from the Upper Hudson River that children under age 15 and women of childbearing age eat no fish from the entire 200 mile stretch of river below Hudson Falls and that the general population eat none of most species of fish caught between the federal dam at Troy and Catskill approximately 40 miles south of Troy The Lower Hudson remains closed to commercial fishing for striped bass and eight other species

The fishing bans and advisories were issued because ingestion of fish is the major potential pathway for exposure for the general population Many previous studies have shown relationships between consumption of fish from contaminated waters and human PCB levels Another potential pathway for exposure less frequently studied to date is inhalation of PCBs NYS DOH and ATSDR launched ldquoThe Hudson River Communities Projectrdquo in 2000 to address this issue (More information on this project is provided below)

POPS and PCBs

The category ldquopersistent organic pollutantsrdquo (POPs) includes man-made chemicals that are by definition persistent in the environment Examples of POPs are polychlorinated biphenyls

5

(PCBs) polychlorinated dibenzo-p-dioxins (PCDDs) polychorinated dibenso-furans (PCDFs) and pprsquo-diphenyldichloroethene (DDE) a breakdown product of the pesticide dichlorodiphenyltrichloroethane (DDT) These chemicals are stored in human fat tissues accumulating over an individualrsquos lifetime They also accumulate in fish and other species thereby concentrating up the food chain Levels of most POPs measured in blood serum of Americans have decreased in the last few decades An exception is polybrominated diphenyl ethers (PBDEs) levels of which have been increasing in recent years While levels of many POPs are declining a large number continue to be detectable in blood for most of the US population (USDHHS 2009)

This follow-up investigation focuses on the Hudson River where the POPs of concern are PCBs PCBs are a group of synthetic chlorinated compounds manufactured in the US through 1977 Production was halted due to concerns about the persistence of PCBs in the environment and the toxicity of PCBs shown in animal models PCBs were used primarily by the electrical industry for capacitors and transformers and were also used in hydraulic fluids fluorescent light fixtures flame retardants inks adhesives carbonless copy paper paints pesticide extenders plasticizers and wire insulators PCBs can be oily liquids or solids and some are volatile and can exist as vapor in air PCBs in the environment are generally mixtures of the many different types known as congeners (ATSDR 2000)

The variety of types of PCBs adds to the complexity of studying adverse animal and human health effects Increasingly PCB research focuses on specific PCB congeners and groups of congeners because specific types of congeners appear to have different mechanisms of action and toxicity Dioxin-like PCBs (DLPCBs) are an important subset of PCBs due to concerns about their toxicity The focus on dioxin-like PCBs is paralleled by increasing research on the group of POPs with dioxin-like properties This type of PCB and POP research assigns toxic potency values relative to dioxin to various PCBs and POPs to provide an overall toxic equivalency value to specific PCB and other POPs

From 2002 through 2007 NYS DOH and SUNY Albany School of Public Health researchers conducted the Hudson River Communities Study among residents of Hudson Falls and Fort Edward the villages where the PCB discharges from two GE electrical capacitor plants occurred The comparison community was upriver from the discharges in Glens Falls This studyrsquos evaluation of exposures included measuring PCBs in outdoor air and in blood serum The study found that PCB levels in outdoor air were statistically significantly higher in the Hudson FallsFort Edward area (072 nanograms per cubic meter) than in the upriver Glens Falls area (040 nanograms per cubic meter) but that the PCB concentrations in the Hudson FallsFort Edward area were lower than those in other communities with known PCB-contaminated sites and similar to levels reported in other locations in the northeastern United States A statistically significant reduction in the total PCB concentrations in outdoor air samples was observed at a distance of 1200 meters (34 mile) from the river from 0760 ngm3 to 0497 ngm3 (Fitzgerald et al 2007 Palmer et al 2008)

The average PCB concentrations in serum for the study and comparison populations did not differ significantly (307 ppb wet weight versus 323 ppb) The average toxic equivalency quantities (TEQ) for dioxin-like PCB congeners for the study and comparison area populations

6

also did not differ In addition the study found no detectable differences in study subjectsrsquo serum PCB levels associated with proximity to contaminated portions of the river or other PCB sources The study did find that lifetime consumption of Hudson River fish was associated with higher serum PCB levels (Fitzgerald 2007 Palmer 2008) Another recent biomonitoring study of Brooklyn Manhattan and New Jersey fisherman who consumed versus those who did not consume their catch from the lower Hudson River and New YorkNew Jersey harbor suggests eating fish from these waters is not associated with an increased body burden of PCBs or most other organochlorines This study did not address the inhalation pathway (Morland 2008)

PCBs and Cardiovascular Disease Outcomes

Research studies on adverse health effects associated with PCB exposure have focused on many types of outcomes organs and biological systems including cancer liver reproductive and developmental effects thyroid endocrine immune system and neurological effects as well as specific effects on the skin and eyes Relatively little research has focused to date on associations between PCB exposures and CVD outcomes The majority of human studies have assessed health effects in occupational groups having relatively high-level exposures among people consuming fish from contaminated waters or people exposed to relatively high levels due to accidental contamination of food (USDHHS NIH 2010)

PCBs have been shown to produce a variety of toxic effects in animals These effects include cancer allergies hypersensitivity damage to the central and peripheral nervous systems reproductive disorders and disruption of the immune system (ATSDR 2000) In contrast research to date does not tell us whether specific types of human health effects have occurred from the low-level exposures experienced by the general population Some human epidemiological studies of relatively low levels of PCDDs PCDFs and PCBs suggest possible relationships not only with neurodevelopment of infants but also diabetes and a condition called insulin resistance conditions that increase CVD risk (Longnecker 2001 Carpenter 2006 Arisawa 2005)

In 1998 a study of a population in Seveso Italy exposed to relatively pure TCDD (dioxin) from an industrial accident showed increased mortality from CVD respiratory diseases and diabetes (Bertazzi 1998) These findings suggested the possibility that other dioxin-like chemicals such as dioxin-like PCBs might increase risk for these diseases Some more recent animal studies described below suggest possible mechanisms of action for PCB effects that increase risk for CVD diabetes and hypertension Another relevant source of information is from research using the biological monitoring data from the National Health and Nutrition Examination Survey (NHANES) project to look at the association between PCB levels in blood and self-reported diabetes and heart disease The body of evidence based on analytical human epidemiological studies remains relatively sparse however

Some recent studies using NHANES biomonitoring data show associations between PCB levels and CVD (Lee 2006 Lee 2007 Ha 2007 Everett 2008) These studies are cross-sectional (comparisons at one point in time) evaluations of PCB levels measured in blood samples from individuals and self-reported CVD and diabetes a risk factor for CVD While these studies use individual-level biomonitoring and self-reported health outcome data they are considered to be

7

relatively weak in terms of providing evidence of causality because both exposure and outcome are measured at the same point in time Given the nature of NHANES data collection and reporting practices these studies are not able to control for possible geographical confounding which may occur if urban versus rural residence is associated with higher PCB levels and higher risk for CVD for example The higher PCB levels among African Americans in these data suggest urban residence may be playing a role in the PCB data

The results of animal studies suggest multiple mechanisms by which PCB exposures could affect cardiovascular risk as well as multiple types of adverse health outcomes One hypothesized mechanisms of action specifically for dioxin-like PCBs is that they may have anti-estrogenic effects which impact CVD risk factors such as serum cholesterol levels and blood pressure (Lind 2004) Several studies suggest PCBs interact with a specific receptor (AhR) and increase cellular oxidative stress causing inflammation cell injury and cardiovascular dysfunction (Hennig 2002) Similarly some studies suggest PCB-induced inflammation of fat cells leads to obesity and increased risk for coronary artery disease (atherosclerosis) (Arsenescu 2008)

3 FOLLOW-UP STUDY DATA SOURCES AND METHODS

This follow-up investigation will provide perspective on the 2004-2005 findings (Sergeev and Carpenter 2005 Shcherbatykh et al 2005) by producing results from analyses conducted at a more detailed level of geography the block group rather than ZIP code in order to better control for confounding When the term confounding is used by epidemioogists it refers to the problem that can arise when there is a factor such as income level that is not the factor of interest in the study but that is related to the risk factor being studied (residential location ndash proxy for exposure) and to the disease being studied (CVD) This type of factor is labeled a ldquoconfounderrdquo because it can confound ie interfere with or confuse the evaluation of associations among the variables of interest in the data All types of epidemiological studies face issues of confounding but ecological ie group-level analyses face additional challenges related to potential confounding because analyses can not account for some important individual differences within the groups that are the units of analysis (Morganstern 1998b Hertz-Picciotto 1998) This investigation was able to control for each individualrsquos age sex and race but not for individual-level risk factors such as income or education level

The follow-up investigationrsquos specific objectives are to

improve the ability to control for confounding by socio-economic status through the use of block groups rather than ZIP codes as the unit of analysis

compare the hospitalization rates among residents of block groups nearest the Hudson River to those among residents of block groups farther from the Hudson River

compare the findings of this follow-up study to those of the earlier studies

Study Area and Exposure Areas

The overall study area is comprised of the 12 counties with geographic areas abutting the Hudson River south of the source of PCB contamination from the General Electric facility in Hudson Falls The 12 counties are Westchester Rockland Putnam Orange Dutchess Ulster

8

Columbia Greene Rensselaer Albany Washington and Saratoga US Census boundaries are used to define the study area because census population estimates provide the data for the study area populationrsquos size and the populationrsquos age sex and race distribution Because the study period spanned 1990 through 2005 and populations change over time 1990 and 2000 Census data were used to linearly interpolate the between-year populations from 1991 through 1999 for each block group 2000 data were used without extrapolation to estimate block group populations from 2001 through 2005

Because block group boundaries change between the Census years 1990 block populations were normalized to fit 2000 block group boundaries before the estimation of sex age and race group specific populations In the digital map 2000 block group boundaries were overlaid on the 1990 block centroids to assign 1990 block population counts to the corresponding 2000 block group boundaries Then 1990 block populations were summed to fit the block group boundaries Next population counts by age group (0-14 15-24 25-34 35-44 45-54 55-64 65-74 75+) sex (male female) and race (white black asian other) were estimated for each year at the block group level This investigation includes all ages over 25 years and the study population is divided into four race categories white black or African-American Asian and ldquoOtherrdquo The Pacific Island group was combined with the Asian group and American Indian ldquootherrdquo and ldquomultiple racesrdquo categories are combined into a larger ldquootherrdquo category

Classification of the population into potential exposure categories for this study is based exclusively on location relative to the Hudson River Two types of exposure proxy categories were developed block group adjacency and block group centroid distance to the river shore For the adjacency categories block groups were assigned to the adjacent category if they directly abutted the Hudson River and all other block groups in the 12 counties were assigned to the nonshyadjacent category

For the distance categories a central point (centroid) within each block group was identified and used to assign block group centroid distance to the river shore The block group centroids were population-weighted which means the geographic distribution of households within 2000 census blocks were used to locate block group centroids based on the geographic population distribution By locating a population-weighted block centroid rather than a more straightforward distance centroid the assignment of block group distance to the river shore more accurately reflects distance to the river for the majority of block group residents Buffers of one-half mile and one mile from the Hudson River boundary were created for the assignment of each block group into one of the three distance categories less than frac12 mile (closest) frac12 to one mile (close) and farther than one mile from the river (distant)

Distance categories were selected a priori at the start of the study Because environmental sampling data were not collected along the length of the river included in this investigation we are not able to determine whether these distance categories represent actual differences in exposures As discussed previously a statistically significant reduction in the total PCB concentrations in outdoor air samples was observed at a distance of 1200 meters (34 mile) from the river from 0760 ngm3 to 0497 ngm3 in a study conducted in the Hudson Falls Fort Edward area near the source of PCBs (Fitzgerald 2007 Palmer 2007)

9

For the adjacency categories of the 2371 block groups in the study area 255 (11) were classified as adjacent to the Hudson River and 2116 (89) were classified as non-adjacent (Figure 2) For the distance categories 225 (9) were classified as closest to the river 216 (9) as close and 1930 block groups (81) as distant from the river

Comparison of Figures 1 and 2 show that block groups include smaller geographic areas than ZIP codes and that block groups are relatively smaller in area in urban versus rural areas The study area includes a total of 368 ZIP codes compared to a total of 2371 block groups ZIP code average population in the study area is approximately 8000 compared to block group average population of 1200 Comparison of Figures 2 and 3 shows that using population-weighted centroid distance to the river produces different exposure areas than adjacency Fifty-three (20) of the adjacent block groups were categorized for the distance indicator as distant because the population in the block group was concentrated in the area farther than one mile from the river On the other hand about 11 of the 2116 non-adjacent block groups were classified as closest (77) or close (162)

Hospitalization Data

Hospitalizations for ischemic heart disease (IHD) and cerebrovascular disease (stroke) were included in this study More specifically hospital discharge records that listed as the principal diagnosis either IHD (ICD-9-CM 410-414 and 4297) or cerebrovascular disease (ICD-9-CM 430-436) and subsets of each of these categories similar to the subsets evaluated in the Sergeev et al and Shcherbatykh studies were included The subsets are AMI and angina subsets of IHD and ischemic and hemorrhagic stroke which are subsets of cerebrovascular disease (stroke) (Principal diagnosis represents the illness for which the person was admitted to the hospital) Angina is an outcome not evaluated in the prior studies It was added here for comparative purposes because it represents an outcome appropriately addressed by primary outpatient rather than inpatient hospital care Angina is one of a group of outcomes included in the recently developed national ldquoPrevention Quality Indicatorsrdquo (USDHHS AHRQ 2007) as an indicator for geographic areas or populations for whom primary care is inadequate Table 3 lists the specific ICD-9-CM codes and definitions included in this investigation

While one of the previously published studies (Huang et al 2006) showed elevations in hypertension (high blood pressure) rates this outcome is rarely listed as the principal reason for hospitalization As a secondary diagnosis however hypertension is extremely common with substantial overlap with the primary diagnoses of IHD and stroke In addition hypertension hospitalizations were not as elevated in the ZIP codes adjacent to the Hudson River as the IHD and stroke outcomes As a practical matter geocoding the hypertension hospitalization data (1364000 hospitalizations) would overwhelm the resources for this project For these reasons hypertension was not included in this follow-up investigation

The source of the hospitalization data was the NYS DOH Statewide Planning and Research Cooperative System (SPARCS) established in 1979 to collect detailed records on discharges from hospitals located in New York State Persons seen in the emergency department but not admitted are not included in the SPARCS data Hospitalizations at US Veterans Administration

10

Hospitals are also not included in this dataset We identified the hospital transfers in the data and excluded them so that individuals are only counted once for the same event

The NYS DOH Data Protection Review Board for SPARCS hospitalization data approved our use of hospitalization records for this investigation and provided approximately 460000 records with a principal diagnosis of either IHD or stroke Because of the large number of cardiovascular hospitalizations and this studyrsquos requirement for geocoding each hospitalized personrsquos address we selected a 50 random sample of the hospital discharges with principal diagnoses of IHD or stroke for inclusion in the study (229948 records) Approximately one percent of these records (2737) were eliminated from the study because they contained no residential address resulting in 227211 hospitalization records 24939 of these records were eliminated after geocoding because they were not within the 12 county study area This left 202272 records remaining in the study

For geocoding to the block group level the hospitalization street addresses were modified as needed using US Postal Service standards The addresses were assigned geographic coordinates using commercially available geocoding software (MapMarker 2004) A combination of land parcel data and street centerline files were used to assign geographic coordinates No contact was made with cases parents legal guardians or next of kin of cases to determine residential locations

Once geographic coordinates were assigned to cases through address-matching the case locations were overlaid onto digital maps of the study area using a geographic information system so that the number of observed cases falling within the study area boundaries could be determined and a 2000 Census block group could be assigned to each hospitalization In order to protect confidentiality no maps of individual case locations are provided Geographic coordinates were not able to be assigned for some addresses for example rural routes or post office boxes Internet searches were conducted for addresses that were located in facilities such as nursing homes and trailer parks

While fewer than one percent of records were missing sex or age information approximately 15000 records seven percent were missing race information Evaluation of the geographic and temporal distribution of these records showed they were not randomly geographically distributed 47 of the records with no race information were from one hospital for that hospital race was missing from 100 of the records in all years except one after 1997 mid-way through the study time-frame Evaluation of other records missing race showed similar patterns of 100 of records missing race in specific years In order to prevent biased hospitalization rate estimates records missing race were assigned race randomly assuming a race distribution equal to that of the individualrsquos residential block group

Statistical Analysis

This investigation used negative binomial multivariable regression models the statistical methods used in the previous studies (Sergeev and Carpenter 2005 [IHD] Shcherbatykh et al 2005 [stroke]) The follow-up studyrsquos analyses compared hospitalization rates for populations in block groups adjacent or in close proximity to the Hudson River to rates for populations more

11

distant from the Hudson River while adjusting for the potential influence of other factors Sex age and race-specific hospitalization count data comprise the dependent variable and the statistical methods attempt to predict the counts using the sex age and race distributions from Census data as well as block group-level demographics such as median income Models were first estimated using basic Census information for age sex and race Then more fully adjusted models were estimated using information at the block group level for income education population density Hispanic ethnicity and distance to the nearest hospital

More specifically each hospitalized personrsquos block group was used to assign quartile categories of block group median household income of population with less than a college education population density of the population identified as Hispanic and distance from the block group centroid to the nearest hospital These variables were selected in advance The population density variable was selected to account for potential effects on hospitalization rates of large differences in population density from the relatively rural northern counties to the more urban southern counties in the study area of the population identified as Hispanic was included because Hispanic ethnicity may be associated with CVD risk but Hispanic ethnicity was not available in the hospitalization records Distance to the hospital was included because studies show that shorter distances increase the likelihood that individuals with AMI and stroke survive to be admitted to the hospital

As described previously the literature about risk factors for CVD shows strong associations between lower income and education and higher CVD risk Regarding the effects of income and education on hospitalization rates the literature from analytic studies specifically of hospitalizations for CVD is quite sparse Most analytic studies focus on prevalence incidence or mortality in population-based studies or cohort studies and there are not many of these An example of a study that was useful as background is a study of income level and CVD mortality that showed strong associations of lower income and higher CVD mortality for men and women with the relationship for men being linear throughout the range of income levels while for women the relationship was only strong at levels of income below the median (Rehkopf 2008)

This follow-up studyrsquos approach was to use the same set of variables for the more fully adjusted models using quartile cut points for all the sexrace groups and types of CVD diagnostic subsets The 2000 Census data were used to create the population-weighted quartile categories to be used in the statistical analyses The quartile cut points are shown below

population density(persons per square mile)

median 1999 household income

with less than a college education

with Hispanic ethnicity

Hospital distance

Quartile 1 61 - lt507 $2499-lt$40968 0 - lt31 0 - lt2 lt25 (median) Quartile 2 507 - lt102 $40968 - lt54650 31 - lt43 2 - lt5 25 - lt6 (50-75) Quartile 3 102 - lt6410 $54650 - lt72944 43 - lt55 5 - lt10 6 ndash 11 (75-90) Quartile 4 6410 - 94372 $72944 - 200001 55 - 100 10 - 86 11 ndash 45 (90-100)

SAS version 9 provided the statistical programs used to estimate hospitalization rates for exposure proxy categories and demographic categories The models provide ldquocontrast

12

estimatesrdquo ie rate ratios for one category versus a reference category of a variable and 95 confidence intervals (CI) for these rate ratios For the location categories rate ratios for the areas adjacent are compared to areas not adjacent to the river (reference) and areas closest and close to the river are compared to areas distant from the river (reference) If a rate ratio is greater than 100 the model has estimated an elevation of hospitalizations in that category of the study population compared to the reference population If the rate ratio is less than 100 then there are fewer than expected hospitalizations compared to the reference population

The magnitude of the elevation or deficit is also estimated by the rate ratio For instance if the hospitalization rate is twice as high in the adjacent population it would result in a rate ratio of 200 while a 50 higher rate would result in a rate ratio of 150 If the hospitalization rate in the comparison group was only half as big as in the reference group this would result in a rate ratio of 050 The 95 confidence interval assists with determining whether the observed differences are statistically meaningful ie statistically significant The confidence interval provides the range in which there is a 95 probability of including the true rate ratio

In addition to directly assess potential confounding we examined the associations between income and exposure proxy categories and between income and CVD hospitalization risk Because the modeling process provides rate ratios that provide comparisons among groups but not the rates themselves we also calculated the rates per population for each sex race and income category in order to directly observe the relative magnitudes of CVD hospitalization rates Finally analyses stratified by income quartile were also conducted in order to assess the potential role of confounding by income The stratified analyses consist of negative binomial regressions conducted separately for each quartile of income

4 RESULTS

Demographics Associations with Proximity to the River

Study Population Table 4 provides detailed demographic characteristics for the total study population and the populations of the adjacent versus non-adjacent and nearer versus farther from the river block groups from the 2000 Census The total population of the study area for this one year of the study time period is 2924631 persons The first row of Table 4 shows that approximately 11 percent of the study area population is classified as living in block groups adjacent to the Hudson River and 89 live in non-adjacent block groups Approximately 8 of the studyrsquos population lives in the closest block groups approximately 10 lives in close block groups and approximately 82 in distant block groups The table also shows that approximately 81 of the total study population is identified as white 9 black 3 Asian less than 01 Pacific Islander 02 American Indian 4 ldquootherrdquo and 2 ldquomultiple racesrdquo

The 2000 Census data (Table 4) also show that the populations in block groups adjacent and nonshyadjacent to the river are similar in terms of the percent of minority population but the adjacent block groups include a smaller percentage of persons who have attended college and a higher percentage of households below the poverty level Median income is $47320 for block groups adjacent versus $62668 for block groups not adjacent to the river Data for the study area from the 1990 Census show similar distributions (data not shown)

13

For the distance to the river categories there is a gradient for the three levels of distance from the river with the closest block groups on average having higher percentages of minorities smaller percentages of persons who have attended college and higher percentages of households below the poverty level (Table 4) The populations of the close and closest block groups are more similar to each other than to the populations of the distant block groups The population of the closest block groups is estimated to be 19 black close block groups 14 black and distant block groups 8 black The proportions for the ldquootherrdquo category used in the analyses which includes multiple races and American Indians are approximately 12 in both the closest block groups and in the close block groups compared to 5 in the block groups most distant from the river Median income for the closest block groups is $40748 close block groups $44962 and most distant block groups $64847

The bar chart below illustrates the correlations between US Census 2000 indicators of block group socio-economic status and block group centroid (population weighted) distance from the river for the study population

2000 Census Demographics for Study Population by Distance to River Category

black other hispanic no college below poverty

of pop ulati on

0

10

20

30

40

50

60

lt12 mile

12 to 1 mile

gt1 mile

Table 5 shows selected demographic information for the person-year estimates used in the study analyses for the 1990-2005 study period Person-year data represent individuals from Census estimates counted repeatedly for each year of the study timeframe As described in the Methods section the estimation of person years used interpolation to account for population changes from 1990-2000 and techniques for taking account of block boundary changes The person-years estimates for demographic categories (complete data not shown) show similar patterns as the 1990 Census data (data not shown) and 2000 Census data (Table 5) Table 5 for example shows nearly identical percentages of population in each of the river distance categories as Table 4

Table 5 also shows person year information for income quartiles and exposure proxy categories as well as race and exposure proxy categories that is consistent with the Table 4 data for the

14

2000 population Table 5 row percentages show for example that while approximately 11 of the total study population live in adjacent block groups approximately 19 of the population in the lowest income quartile live in adjacent block groups By design approximately 25 of the total study population is in each of the income quartiles Table 5 column percentages show however that in the adjacent block groups 41 of the population is in the lowest income quartile and only 9 are in the highest income quartile

For the block groups within frac12 mile of the river the distribution among income quartiles is even more skewed towards the lowest income quartile 61 of the population living within frac12 mile of the river is in the lowest income quartile and only 6 is in the highest income quartile Given that approximately 25 of the total study population is within each income quartile these percentages indicate that living within the block groups closest to the river is associated with more than a doubling of the likelihood of being in a block group in the lowest socio-economic status quartile 436 of those living in the block groups close to the river (12 to one mile from the river) are in block groups in the lowest income quartile While less than a doubling this indicates that living in the close block groups is associated with a 75 increase in the likelihood of being in the lowest quartile for socio-economic status

Hospitalizations Table 6 shows the process of removing hospitalization records from the study for a variety of reasons The first column includes the 202272 records that contained a residential address and were located within the 12-county study area as described in the Methods section The next five columns show the records remaining after data cleaning exclusion of transfers elimination of records with missing covariates and exclusions due to poor geocoding results Approximately 33000 records were eliminated because they were transfers from one hospital to another for the same event For approximately 1100 records geocoding was inadequate for assigning block group for imputing race as described in the Methods section 336 records were missing sex (11) or age (325) Of the 167733 records remaining 154220 (92) were able to be adequately geocoded to a street address for assignment in a specific block group While approximately eight percent of hospitalizations were eliminated due to inadequate geocoding Table 6 shows that the percentages of hospitalization records for white black Asian and ldquootherrdquo categories and the distribution among diagnostic categories change very little after these records were dropped

The final column of Table 6 shows that of the 154220 hospitalizations in the analysis 101012 IHD hospitalizations comprise the largest health outcome category and are 65 of the hospitalizations in the study AMI hospitalizations (35811) account for 35 of the IHD hospitalizations and angina (4731) accounts for five percent of the IHD hospitalizations Stroke hospitalizations (53208) comprise 34 of the studyrsquos hospitalizations Approximately 36 of the stroke hospitalizations are for ischemic stroke (19371) and approximately 11 are for hemorrhagic stroke (5995)

Table 7 shows the distributions of hospitalizations among the four income quartiles While each income quartile (by definition) contains approximately 25 of the study population for all types of hospitalizations we see that the lowest income quartile includes more than 25 of hospitalizations from a minimum of 286 (IHD) to a maximum of 356 of hospitalizations (angina) The highest income quartile includes fewer than 25 of hospitalizations for each CVD

15

category ranging from 182 for angina to 224 for hemorrhagic stroke These deviations from the 25 quartile distribution show a strong linear pattern with the largest percentage of hospitalizations coming from residents of block groups in the lowest income quartile and the smallest percentages coming from residents of the most wealthy block groups For AMI the lowest income quartilersquos share of hospitalizations is 50 higher than the highest income quartilersquos share of hospitalizations (30 vs 20) IHD and stroke show similar differences and angina shows a doubling with the lowest income quartile contributing 36 of hospitalizations compared to the highest income quartilersquos 18 share of hospitalizations

Table 7 also shows the distribution of hospitalizations among the race categories and the study population race category distribution for comparison Blacks show lower percentages of IHD and AMI than their population percentages and higher percentages of angina and stroke While full presentation of race and sex findings is beyond the scope of this report the lower than expected percentages of IHD and AMI among blacks who on average have lower socioshyeconomic status and are at greater risk for CVD hospitalizations raised questions that were addressed by calculating hospitalization rates per 1000 population for each sex and race group for each income quartile (Appendix B Table 2) Figures 4 5 and 6 show these rates for the white black and Asian categories (These rates for the ldquootherrdquo category are comparatively very high and are not shown on these graphs This issue is addressed in the Discussion section) Figures 4 5 and 6 show the consistent and strong effects of income quartile on hospitalization rates for AMI and stroke for whites blacks and Asians of both sexes The other outcomes show similar patterns (The ldquootherrdquo group showed an opposite pattern [See Discussion Section])

Demographics Summary

The strong associations of lower income with residence closer to the river (Table 5) and lower income with higher hospitalization rates (Tables 7 amp 8) create a difficult challenge for this investigation The multivariable analyses of hospitalization rates attempt to control for socioshyeconomic differences However because these income differences which strongly affect hospitalization rates are also strongly correlated with the gradations of exposure (river distance) it will be difficult to disentangle the influence of socio-economic status from the influence of potential exposures

Statistical Analyses Proximity to the River and CVD Hospitalization Risk

All the variables described above were included in the multivariable regression models and almost every quartile category for every variable showed statistically significant and expected types of effects for all the CVD hospitalization categories (Appendix B Tables 1a and 1b) The following presentation of results focuses on the rate ratios for the exposure proxy variables Table 8a (IHD) and 8b (stroke) show the regression modeling results Results are shown first for models that include only age sex and race in addition to the exposure proxy categories and second for models that also include median income education population density Hispanic ethnicity and distance to nearest hospital The tables provide numbers of hospitalizations and person-years to assist with interpretation For example for total IHD we can see that while 108 of the study population lives in adjacent block groups 122 of the hospitalizations are among people in adjacent block groups This difference of 14 percentage points is the basis for

16

the finding of a statistically significantly elevated rate ratio (risk) for IHD among people living in the adjacent block groups (RR=133 CI 112-157)

Looking across the columns of Tables 8a and 8b we can see the rate ratios for each of the outcomes for the adjacent versus non-adjacent block groups and the results are similar to those for IHD The elevated rate ratios range from 123 for angina to 136 for AMI The only rate ratio not elevated in these first models adjusting only for age sex and race is for hemorrhagic stroke Going down the rows of the Table the next set of rate ratios are for the distance to the river categories The findings are very similar to those for the adjacent block groups Here however we have two exposure proxy categories allowing comparison of the two These rate ratios show no evidence that living in the block groups closest to the river (lt12 mile to the river) produces greater risk than living in the block groups close to the river (12 to 1 mile from the river)

The next set of rate ratios are for models with additional adjustment for the demographics of each hospitalized personrsquos block group as well as distance to the nearest hospital As expected adjustment for block group median income and other demographics accounts for some of the variability in hospitalization counts and thereby reduces the magnitude of the rate ratios estimated for the river adjacency and distance categories The rate ratio elevation for total IHD for example is reduced from 133 to 107 (RR=107 CI 104-111) The elevated rate ratios in these more fully adjusted models for the adjacency analyses are of similarly small magnitude ranging from 106 for AMI to 110 for ischemic stroke Four of the six outcomes show statistically significant elevations Again only hemorrhagic stroke shows no evidence of an elevated rate ratio

For the river distance categories none of the rate ratios show any evidence that living in the block groups closest to the river (lt12 mile to the river) produces greater risk than living in the block groups close to the river (12 to 1 mile) Of the six types of outcomes only two show statistically significantly elevated rate ratios for living in the closest block groups (total stroke and ischemic stroke) For the close block groups four out of six outcomes show statistically significantly elevated rate ratios The statistically significantly elevated rate ratios vary from 106 (AMI stroke) to 113 (angina)

To address the strong evidence described earlier for confounding in these data ie the association of lower income with both CVD hospitalizations and residence closer to the river we also conducted the analyses separately for each quartile of income These results are presented in Table 9 The first results are from analyses adjusting only for age sex and race and the second set of results are from models including the additional demographic factors For simplicity rate ratios are shown only for block groups closest to the river compared as in the other models to block groups farther than one mile from the river (Results were similar for the other distance categories (data not shown)

For the two lowest income quartiles Table 9 shows results similar to those for the study population as a whole (Table 8a 8b) However for the two highest income quartiles which represent frac12 of the study population Table 9 shows a consistent pattern of reduced rate ratios for residence in the closest block groups within frac12 mile of the river In the highest income quartile

17

economic status are likely not the cause for the elevated hospitalization rates in these Hudson River area ZIP codes

In these previously published reports Behavioral Risk Factor Surveillance System (BRFSS) data for entire counties are provided to support statements about the population of Hudson River ZIP codes having higher income and healthier lifestyles However because the ZIP codes adjacent to the Hudson River include only a proportion of the population of the counties abutting the river these county-level BRFSS data may be misleading regarding characteristics of the ZIP codes adjacent to the river (Figure 1) In fact data from the US Census show that the ZIP codes adjacent to the river have lower median income than the counties within which they are located The median income for the total population of the twelve counties adjacent to the Hudson River was $61010 in 2000 while the median income for the ZIP codes adjacent to the river was $45356 This median income for the population of the ZIP codes adjacent to the river is also slightly lower than the statewide (excluding NYC) median income of $47735 (US Census 2000) Thus the Hudson River ZIP codes analyzed in the Sergeev and Carpenter Shcherbatykh et al and Huang et al studies include a population of lower not higher socio-economic status compared to the rest of NYS (excluding NYC)

Because CVD and many CVD risk factors are associated with lower socio-economic status it is especially important to adjust as much as possible for socio-economic status when conducting comparative analyses of CVD hospitalizations or disease incidence or prevalence Regarding the hypothesis about POPs and CVD research studies described in more detail below have suggested possible biological mechanisms for POP exposuresrsquo effects on the heart There is very little research on human exposures to POPs and potential cardiovascular effects however

In response to Congressman Hincheyrsquos request for follow-up of the 2004-2005 findings ATSDR and NYS DOH agreed to conduct a similar study also ecological but at a different level of geography the Census block group By using a level of geography that is generally smaller in area and population than a ZIP code this follow-up seeks to more accurately adjust for socioshyeconomic and other factors that affect cardiovascular hospitalizations and to improve upon the geographic classification of relative distance to the Hudson River In addition rather than comparing hospitalizations among residents of Hudson River area ZIP codes to hospitalizations among residents of all other NYS (exclusive of NYC) ZIP codes with no waste sites with POPs this follow-up investigationrsquos analyses are restricted to the twelve counties abutting the Hudson river from Washington County south to Westchester County Within the study arearsquos twelve counties cardiovascular hospitalization rates in areas closest to the Hudson River are compared with rates in areas farther from the river This more localized approach may also assist with reducing the potential impact of regional differences in medical care practice or reporting factors that could affect the hospitalization data (Ko et al 2007) (Table 2 shows a more detailed listing of the methods used in the prior studies and follow-up study)

Regarding environmental exposures it is important to note that living near a hazardous waste site that contains PCBs or other POPs does not necessarily mean that human exposure from the site has occurred Depending upon the conditions at a site or the specific chemicals exposures may be highly improbable Information on body burden for example PCB levels in blood or on levels of the chemical in off-site air dust or soil is needed to support a claim of exposure The

2

studies published in 2005-2006 did not provide specific information such as environmental or biomonitoring data in support of the assumption that residence in ZIP codes containing hazardous waste sites with POPs or residence in ZIP codes adjacent to the Hudson River resulted in increased exposures to POPs in the population For the purpose of this follow-up we did not conduct environmental or human biological sampling to address this issue Rather this follow-up replicates the prior studiesrsquo approach with some enhancements to better control for confounding particularly confounding by socio-economic status

This descriptive epidemiologic investigation uses already existing hospitalization data from the Statewide Planning and Research Cooperative System and population data from the United States Census to evaluate comparative rates of cardiovascular hospitalizations While this type of investigation can not show proof of cause and effect it can assist in interpreting relationships among indicators of CVD burden and hypothesized exposures associated with distance from the Hudson River This descriptive investigation may also assist in determining whether additional health investigations of these associations are warranted

2 BACKGROUND

Cardiovascular Disease

The widely known risk factors for cardiovascular disease (CVD) in addition to age sex race and family history include tobacco use high blood pressure high serum cholesterol alcohol use obesity low fruit and vegetable intake physical inactivity diabetes low socio-economic status mental ill-health psychosocial stress and the use of certain medications (WHO 2010 NYS DOH 2009 Wing 1988 NYS DOH 2009a)

CVD Definitions CVD is a general term for any disease of the circulatory system The two main types of CVD are ischemic heart disease also termed coronary heart disease and cerebrovascular disease also called stroke Ischemic heart disease (IHD) results from restricted blood flow through the arteries supplying the heart muscle The most serious risk from IHD is acute myocardial infarction (AMI) another term for heart attack AMI occurs when blood flow to the heart is suddenly reduced or stopped usually due to a blockage in a coronary artery Angina is severe constricting pain in the chest often radiating from the area immediately over the heart to the left shoulder and down the arm Cerebrovascular disease refers primarily to stroke the sudden impairment of brain function resulting from interruption of circulation to the brain Ischemic stroke results from blockage of an artery supplying blood to the brain Hemorrhagic stroke occurs due to the escape of blood from a ruptured artery supplying blood to the brain

CVD Burden and Trends CVD is the leading cause of death for men and women of all ages in the United States and in New York State In 1999 for every death from IHD there were five hospitalizations for ischemic disease in NYS (NYS DOH 2009a) Mortality rates from CVD have been declining steadily since the 1980rsquos Research studies estimate that approximately half of the decline is due to improved treatments for CVD and its risk factors such as high cholesterol and high blood pressure and approximately half is due to decreasing incidence of CVD The decrease in incidence is due to trends in health behaviors primarily reduced smoking rates

3

CVD mortality rates do not appear to be declining for all segments of the population however and may actually be increasing among some groups particularly people of lower socio-economic status (Barnett et al 1999) This may be due in part to higher rates of smoking among people of lower socio-economic status as well as to other factors related to socio-economic status and race A growing literature points to multiple causes for CVD and other health disparities including health practices psychosocial stress limited resources discrimination and access to health care (Health United States 2007) Increases in the prevalence of diabetes and obesity in the general population partially offset advances in treatment and favorable trends for other risk factors An increasing ischemic heart disease mortality rate in women ages 35 to 54 since 1997 is attributed to these unfavorable risk factor trends (Ford 2007 Lloyd-Jones 2009) New York State (NYS) and other northeast states have high IHD and low stroke mortality with NYS having the highest IHD mortality and lowest stroke mortality among the 50 states (Howard et al 2009)

CVD Data Sources CVD mortality rates are more easily measured than incidence prevalence or hospitalizations because mortality data from vital records are a comprehensive and relatively accessible data source Information for estimating the incidence and prevalence of CVD outcomes comes primarily from health interview physical examination and record surveys Hospitalization data are available for NYS and the US but are generally collected and analyzed in order to track trends in hospital care and resources expended Actual CVD incidence data validated by physician diagnosis physician examination or reviews of medical records are not directly available for the entire US or for specific geographic areas of NYS Such information is available from population-based cohort studies that follow people over time in selected communities While the concern underlying this research project is to understand CVD levels in the community ie incidence and prevalence we do not have data that directly provide us with this information In the absence of such data hospitalizations are frequently used as indicators of the burden of disease in communities particularly in surveillance or hypothesis-generating investigations such as this one In NYS the Behavioral Risk Factor Surveillance System (BRFSS) an ongoing monthly telephone survey of adults aged 18 years and older provides information about the burden of CVD by age sex race income and education

CVD Risk Factor Data The BRFSS data for NYS show that over 20 of the population aged 65 and older reported having been diagnosed with CVD versus less than five percent of the population under 65 Males were slightly more likely to report having CVD than females Nonshyhispanic whites compared to other ethnic groups combined were slightly more likely to report having CVD None of these slight differences were statistically significant however The only statistically significant demographic risk factors in the BRFSS data were low income and education levels Statistically significant elevations of reported CVD were observed for respondents with incomes below $25000 and respondents with a high school degree or less education (NYS DOH 2009a)

Another factor that can affect the outcome measure hospitalizations is the likelihood of mortality prior to reaching the hospital Studies show a variety of factors affect whether an individual experiencing a heart attack or stroke outside the hospital survives to be counted in hospitalization data These factors include lack of knowledge about personal risk or prior disease socio-economic status race marital status and distance to the hospital (Galea 2007

4

Barnett 2006 Ayala 2003) Population-based studies report sudden cardiac death usually outside the hospital to be the first overt manifestation of heart disease for 40 to 60 of all cases ( Kannel 1987 deVreede-Swagemakers 1997 UK Heart Attack Study 1998) An analysis of out-of-hospital mortality from AMI in Pennsylvania in 1998 showed that in relatively rural areas where travel time to the hospital was 25 minutes or longer 72 of all AMI deaths occurred outside the hospital compared with a statewide average of 49 of AMI deaths occurring outside of the hospital (OrsquoNeill 2003) In New York State 55 of IHD deaths occur outside the hospital compared to 35 of stroke deaths (NYS DOH 2009b)

Hudson River PCB Contamination

During an approximate 30-year period ending in 1977 the General Electric (GE) plants in Hudson Falls and Fort Edward used PCBs in the manufacture of electrical capacitors and discharged large quantities of PCBs into the upper Hudson River Estimates of the total direct discharges to the river are as high as 1330000 pounds In 1983 the US EPA classified a 200shymile section of the Hudson River from Hudson Falls to New York City as a National Priority List site making it one of the largest Superfund sites in the United States In 2002 the US EPA issued a Record of Decision calling for targeted environmental dredging and removal of PCB-contaminated sediment in the upper River (USEPA 2002) The dredging began in May 2009

The Superfund environmental investigations and planned remedial actions have focused on the areas of the river near Hudson Falls and Fort Edward where the highest levels of PCBs in sediments are located The 2002 Record of Decision provides information about the human health risk assessment conducted for the Upper and Mid-Hudson and the ecological risk assessment conducted for the Upper and Lower Hudson River (USEPA 2002) Fish contamination and fish consumption are the primary concerns at the site

In 1976 fishing was banned in the upper river and in 1995 a catch and release policy replaced the ban NYS DOH continues to recommend people eat no fish from the Upper Hudson River that children under age 15 and women of childbearing age eat no fish from the entire 200 mile stretch of river below Hudson Falls and that the general population eat none of most species of fish caught between the federal dam at Troy and Catskill approximately 40 miles south of Troy The Lower Hudson remains closed to commercial fishing for striped bass and eight other species

The fishing bans and advisories were issued because ingestion of fish is the major potential pathway for exposure for the general population Many previous studies have shown relationships between consumption of fish from contaminated waters and human PCB levels Another potential pathway for exposure less frequently studied to date is inhalation of PCBs NYS DOH and ATSDR launched ldquoThe Hudson River Communities Projectrdquo in 2000 to address this issue (More information on this project is provided below)

POPS and PCBs

The category ldquopersistent organic pollutantsrdquo (POPs) includes man-made chemicals that are by definition persistent in the environment Examples of POPs are polychlorinated biphenyls

5

(PCBs) polychlorinated dibenzo-p-dioxins (PCDDs) polychorinated dibenso-furans (PCDFs) and pprsquo-diphenyldichloroethene (DDE) a breakdown product of the pesticide dichlorodiphenyltrichloroethane (DDT) These chemicals are stored in human fat tissues accumulating over an individualrsquos lifetime They also accumulate in fish and other species thereby concentrating up the food chain Levels of most POPs measured in blood serum of Americans have decreased in the last few decades An exception is polybrominated diphenyl ethers (PBDEs) levels of which have been increasing in recent years While levels of many POPs are declining a large number continue to be detectable in blood for most of the US population (USDHHS 2009)

This follow-up investigation focuses on the Hudson River where the POPs of concern are PCBs PCBs are a group of synthetic chlorinated compounds manufactured in the US through 1977 Production was halted due to concerns about the persistence of PCBs in the environment and the toxicity of PCBs shown in animal models PCBs were used primarily by the electrical industry for capacitors and transformers and were also used in hydraulic fluids fluorescent light fixtures flame retardants inks adhesives carbonless copy paper paints pesticide extenders plasticizers and wire insulators PCBs can be oily liquids or solids and some are volatile and can exist as vapor in air PCBs in the environment are generally mixtures of the many different types known as congeners (ATSDR 2000)

The variety of types of PCBs adds to the complexity of studying adverse animal and human health effects Increasingly PCB research focuses on specific PCB congeners and groups of congeners because specific types of congeners appear to have different mechanisms of action and toxicity Dioxin-like PCBs (DLPCBs) are an important subset of PCBs due to concerns about their toxicity The focus on dioxin-like PCBs is paralleled by increasing research on the group of POPs with dioxin-like properties This type of PCB and POP research assigns toxic potency values relative to dioxin to various PCBs and POPs to provide an overall toxic equivalency value to specific PCB and other POPs

From 2002 through 2007 NYS DOH and SUNY Albany School of Public Health researchers conducted the Hudson River Communities Study among residents of Hudson Falls and Fort Edward the villages where the PCB discharges from two GE electrical capacitor plants occurred The comparison community was upriver from the discharges in Glens Falls This studyrsquos evaluation of exposures included measuring PCBs in outdoor air and in blood serum The study found that PCB levels in outdoor air were statistically significantly higher in the Hudson FallsFort Edward area (072 nanograms per cubic meter) than in the upriver Glens Falls area (040 nanograms per cubic meter) but that the PCB concentrations in the Hudson FallsFort Edward area were lower than those in other communities with known PCB-contaminated sites and similar to levels reported in other locations in the northeastern United States A statistically significant reduction in the total PCB concentrations in outdoor air samples was observed at a distance of 1200 meters (34 mile) from the river from 0760 ngm3 to 0497 ngm3 (Fitzgerald et al 2007 Palmer et al 2008)

The average PCB concentrations in serum for the study and comparison populations did not differ significantly (307 ppb wet weight versus 323 ppb) The average toxic equivalency quantities (TEQ) for dioxin-like PCB congeners for the study and comparison area populations

6

also did not differ In addition the study found no detectable differences in study subjectsrsquo serum PCB levels associated with proximity to contaminated portions of the river or other PCB sources The study did find that lifetime consumption of Hudson River fish was associated with higher serum PCB levels (Fitzgerald 2007 Palmer 2008) Another recent biomonitoring study of Brooklyn Manhattan and New Jersey fisherman who consumed versus those who did not consume their catch from the lower Hudson River and New YorkNew Jersey harbor suggests eating fish from these waters is not associated with an increased body burden of PCBs or most other organochlorines This study did not address the inhalation pathway (Morland 2008)

PCBs and Cardiovascular Disease Outcomes

Research studies on adverse health effects associated with PCB exposure have focused on many types of outcomes organs and biological systems including cancer liver reproductive and developmental effects thyroid endocrine immune system and neurological effects as well as specific effects on the skin and eyes Relatively little research has focused to date on associations between PCB exposures and CVD outcomes The majority of human studies have assessed health effects in occupational groups having relatively high-level exposures among people consuming fish from contaminated waters or people exposed to relatively high levels due to accidental contamination of food (USDHHS NIH 2010)

PCBs have been shown to produce a variety of toxic effects in animals These effects include cancer allergies hypersensitivity damage to the central and peripheral nervous systems reproductive disorders and disruption of the immune system (ATSDR 2000) In contrast research to date does not tell us whether specific types of human health effects have occurred from the low-level exposures experienced by the general population Some human epidemiological studies of relatively low levels of PCDDs PCDFs and PCBs suggest possible relationships not only with neurodevelopment of infants but also diabetes and a condition called insulin resistance conditions that increase CVD risk (Longnecker 2001 Carpenter 2006 Arisawa 2005)

In 1998 a study of a population in Seveso Italy exposed to relatively pure TCDD (dioxin) from an industrial accident showed increased mortality from CVD respiratory diseases and diabetes (Bertazzi 1998) These findings suggested the possibility that other dioxin-like chemicals such as dioxin-like PCBs might increase risk for these diseases Some more recent animal studies described below suggest possible mechanisms of action for PCB effects that increase risk for CVD diabetes and hypertension Another relevant source of information is from research using the biological monitoring data from the National Health and Nutrition Examination Survey (NHANES) project to look at the association between PCB levels in blood and self-reported diabetes and heart disease The body of evidence based on analytical human epidemiological studies remains relatively sparse however

Some recent studies using NHANES biomonitoring data show associations between PCB levels and CVD (Lee 2006 Lee 2007 Ha 2007 Everett 2008) These studies are cross-sectional (comparisons at one point in time) evaluations of PCB levels measured in blood samples from individuals and self-reported CVD and diabetes a risk factor for CVD While these studies use individual-level biomonitoring and self-reported health outcome data they are considered to be

7

relatively weak in terms of providing evidence of causality because both exposure and outcome are measured at the same point in time Given the nature of NHANES data collection and reporting practices these studies are not able to control for possible geographical confounding which may occur if urban versus rural residence is associated with higher PCB levels and higher risk for CVD for example The higher PCB levels among African Americans in these data suggest urban residence may be playing a role in the PCB data

The results of animal studies suggest multiple mechanisms by which PCB exposures could affect cardiovascular risk as well as multiple types of adverse health outcomes One hypothesized mechanisms of action specifically for dioxin-like PCBs is that they may have anti-estrogenic effects which impact CVD risk factors such as serum cholesterol levels and blood pressure (Lind 2004) Several studies suggest PCBs interact with a specific receptor (AhR) and increase cellular oxidative stress causing inflammation cell injury and cardiovascular dysfunction (Hennig 2002) Similarly some studies suggest PCB-induced inflammation of fat cells leads to obesity and increased risk for coronary artery disease (atherosclerosis) (Arsenescu 2008)

3 FOLLOW-UP STUDY DATA SOURCES AND METHODS

This follow-up investigation will provide perspective on the 2004-2005 findings (Sergeev and Carpenter 2005 Shcherbatykh et al 2005) by producing results from analyses conducted at a more detailed level of geography the block group rather than ZIP code in order to better control for confounding When the term confounding is used by epidemioogists it refers to the problem that can arise when there is a factor such as income level that is not the factor of interest in the study but that is related to the risk factor being studied (residential location ndash proxy for exposure) and to the disease being studied (CVD) This type of factor is labeled a ldquoconfounderrdquo because it can confound ie interfere with or confuse the evaluation of associations among the variables of interest in the data All types of epidemiological studies face issues of confounding but ecological ie group-level analyses face additional challenges related to potential confounding because analyses can not account for some important individual differences within the groups that are the units of analysis (Morganstern 1998b Hertz-Picciotto 1998) This investigation was able to control for each individualrsquos age sex and race but not for individual-level risk factors such as income or education level

The follow-up investigationrsquos specific objectives are to

improve the ability to control for confounding by socio-economic status through the use of block groups rather than ZIP codes as the unit of analysis

compare the hospitalization rates among residents of block groups nearest the Hudson River to those among residents of block groups farther from the Hudson River

compare the findings of this follow-up study to those of the earlier studies

Study Area and Exposure Areas

The overall study area is comprised of the 12 counties with geographic areas abutting the Hudson River south of the source of PCB contamination from the General Electric facility in Hudson Falls The 12 counties are Westchester Rockland Putnam Orange Dutchess Ulster

8

Columbia Greene Rensselaer Albany Washington and Saratoga US Census boundaries are used to define the study area because census population estimates provide the data for the study area populationrsquos size and the populationrsquos age sex and race distribution Because the study period spanned 1990 through 2005 and populations change over time 1990 and 2000 Census data were used to linearly interpolate the between-year populations from 1991 through 1999 for each block group 2000 data were used without extrapolation to estimate block group populations from 2001 through 2005

Because block group boundaries change between the Census years 1990 block populations were normalized to fit 2000 block group boundaries before the estimation of sex age and race group specific populations In the digital map 2000 block group boundaries were overlaid on the 1990 block centroids to assign 1990 block population counts to the corresponding 2000 block group boundaries Then 1990 block populations were summed to fit the block group boundaries Next population counts by age group (0-14 15-24 25-34 35-44 45-54 55-64 65-74 75+) sex (male female) and race (white black asian other) were estimated for each year at the block group level This investigation includes all ages over 25 years and the study population is divided into four race categories white black or African-American Asian and ldquoOtherrdquo The Pacific Island group was combined with the Asian group and American Indian ldquootherrdquo and ldquomultiple racesrdquo categories are combined into a larger ldquootherrdquo category

Classification of the population into potential exposure categories for this study is based exclusively on location relative to the Hudson River Two types of exposure proxy categories were developed block group adjacency and block group centroid distance to the river shore For the adjacency categories block groups were assigned to the adjacent category if they directly abutted the Hudson River and all other block groups in the 12 counties were assigned to the nonshyadjacent category

For the distance categories a central point (centroid) within each block group was identified and used to assign block group centroid distance to the river shore The block group centroids were population-weighted which means the geographic distribution of households within 2000 census blocks were used to locate block group centroids based on the geographic population distribution By locating a population-weighted block centroid rather than a more straightforward distance centroid the assignment of block group distance to the river shore more accurately reflects distance to the river for the majority of block group residents Buffers of one-half mile and one mile from the Hudson River boundary were created for the assignment of each block group into one of the three distance categories less than frac12 mile (closest) frac12 to one mile (close) and farther than one mile from the river (distant)

Distance categories were selected a priori at the start of the study Because environmental sampling data were not collected along the length of the river included in this investigation we are not able to determine whether these distance categories represent actual differences in exposures As discussed previously a statistically significant reduction in the total PCB concentrations in outdoor air samples was observed at a distance of 1200 meters (34 mile) from the river from 0760 ngm3 to 0497 ngm3 in a study conducted in the Hudson Falls Fort Edward area near the source of PCBs (Fitzgerald 2007 Palmer 2007)

9

For the adjacency categories of the 2371 block groups in the study area 255 (11) were classified as adjacent to the Hudson River and 2116 (89) were classified as non-adjacent (Figure 2) For the distance categories 225 (9) were classified as closest to the river 216 (9) as close and 1930 block groups (81) as distant from the river

Comparison of Figures 1 and 2 show that block groups include smaller geographic areas than ZIP codes and that block groups are relatively smaller in area in urban versus rural areas The study area includes a total of 368 ZIP codes compared to a total of 2371 block groups ZIP code average population in the study area is approximately 8000 compared to block group average population of 1200 Comparison of Figures 2 and 3 shows that using population-weighted centroid distance to the river produces different exposure areas than adjacency Fifty-three (20) of the adjacent block groups were categorized for the distance indicator as distant because the population in the block group was concentrated in the area farther than one mile from the river On the other hand about 11 of the 2116 non-adjacent block groups were classified as closest (77) or close (162)

Hospitalization Data

Hospitalizations for ischemic heart disease (IHD) and cerebrovascular disease (stroke) were included in this study More specifically hospital discharge records that listed as the principal diagnosis either IHD (ICD-9-CM 410-414 and 4297) or cerebrovascular disease (ICD-9-CM 430-436) and subsets of each of these categories similar to the subsets evaluated in the Sergeev et al and Shcherbatykh studies were included The subsets are AMI and angina subsets of IHD and ischemic and hemorrhagic stroke which are subsets of cerebrovascular disease (stroke) (Principal diagnosis represents the illness for which the person was admitted to the hospital) Angina is an outcome not evaluated in the prior studies It was added here for comparative purposes because it represents an outcome appropriately addressed by primary outpatient rather than inpatient hospital care Angina is one of a group of outcomes included in the recently developed national ldquoPrevention Quality Indicatorsrdquo (USDHHS AHRQ 2007) as an indicator for geographic areas or populations for whom primary care is inadequate Table 3 lists the specific ICD-9-CM codes and definitions included in this investigation

While one of the previously published studies (Huang et al 2006) showed elevations in hypertension (high blood pressure) rates this outcome is rarely listed as the principal reason for hospitalization As a secondary diagnosis however hypertension is extremely common with substantial overlap with the primary diagnoses of IHD and stroke In addition hypertension hospitalizations were not as elevated in the ZIP codes adjacent to the Hudson River as the IHD and stroke outcomes As a practical matter geocoding the hypertension hospitalization data (1364000 hospitalizations) would overwhelm the resources for this project For these reasons hypertension was not included in this follow-up investigation

The source of the hospitalization data was the NYS DOH Statewide Planning and Research Cooperative System (SPARCS) established in 1979 to collect detailed records on discharges from hospitals located in New York State Persons seen in the emergency department but not admitted are not included in the SPARCS data Hospitalizations at US Veterans Administration

10

Hospitals are also not included in this dataset We identified the hospital transfers in the data and excluded them so that individuals are only counted once for the same event

The NYS DOH Data Protection Review Board for SPARCS hospitalization data approved our use of hospitalization records for this investigation and provided approximately 460000 records with a principal diagnosis of either IHD or stroke Because of the large number of cardiovascular hospitalizations and this studyrsquos requirement for geocoding each hospitalized personrsquos address we selected a 50 random sample of the hospital discharges with principal diagnoses of IHD or stroke for inclusion in the study (229948 records) Approximately one percent of these records (2737) were eliminated from the study because they contained no residential address resulting in 227211 hospitalization records 24939 of these records were eliminated after geocoding because they were not within the 12 county study area This left 202272 records remaining in the study

For geocoding to the block group level the hospitalization street addresses were modified as needed using US Postal Service standards The addresses were assigned geographic coordinates using commercially available geocoding software (MapMarker 2004) A combination of land parcel data and street centerline files were used to assign geographic coordinates No contact was made with cases parents legal guardians or next of kin of cases to determine residential locations

Once geographic coordinates were assigned to cases through address-matching the case locations were overlaid onto digital maps of the study area using a geographic information system so that the number of observed cases falling within the study area boundaries could be determined and a 2000 Census block group could be assigned to each hospitalization In order to protect confidentiality no maps of individual case locations are provided Geographic coordinates were not able to be assigned for some addresses for example rural routes or post office boxes Internet searches were conducted for addresses that were located in facilities such as nursing homes and trailer parks

While fewer than one percent of records were missing sex or age information approximately 15000 records seven percent were missing race information Evaluation of the geographic and temporal distribution of these records showed they were not randomly geographically distributed 47 of the records with no race information were from one hospital for that hospital race was missing from 100 of the records in all years except one after 1997 mid-way through the study time-frame Evaluation of other records missing race showed similar patterns of 100 of records missing race in specific years In order to prevent biased hospitalization rate estimates records missing race were assigned race randomly assuming a race distribution equal to that of the individualrsquos residential block group

Statistical Analysis

This investigation used negative binomial multivariable regression models the statistical methods used in the previous studies (Sergeev and Carpenter 2005 [IHD] Shcherbatykh et al 2005 [stroke]) The follow-up studyrsquos analyses compared hospitalization rates for populations in block groups adjacent or in close proximity to the Hudson River to rates for populations more

11

distant from the Hudson River while adjusting for the potential influence of other factors Sex age and race-specific hospitalization count data comprise the dependent variable and the statistical methods attempt to predict the counts using the sex age and race distributions from Census data as well as block group-level demographics such as median income Models were first estimated using basic Census information for age sex and race Then more fully adjusted models were estimated using information at the block group level for income education population density Hispanic ethnicity and distance to the nearest hospital

More specifically each hospitalized personrsquos block group was used to assign quartile categories of block group median household income of population with less than a college education population density of the population identified as Hispanic and distance from the block group centroid to the nearest hospital These variables were selected in advance The population density variable was selected to account for potential effects on hospitalization rates of large differences in population density from the relatively rural northern counties to the more urban southern counties in the study area of the population identified as Hispanic was included because Hispanic ethnicity may be associated with CVD risk but Hispanic ethnicity was not available in the hospitalization records Distance to the hospital was included because studies show that shorter distances increase the likelihood that individuals with AMI and stroke survive to be admitted to the hospital

As described previously the literature about risk factors for CVD shows strong associations between lower income and education and higher CVD risk Regarding the effects of income and education on hospitalization rates the literature from analytic studies specifically of hospitalizations for CVD is quite sparse Most analytic studies focus on prevalence incidence or mortality in population-based studies or cohort studies and there are not many of these An example of a study that was useful as background is a study of income level and CVD mortality that showed strong associations of lower income and higher CVD mortality for men and women with the relationship for men being linear throughout the range of income levels while for women the relationship was only strong at levels of income below the median (Rehkopf 2008)

This follow-up studyrsquos approach was to use the same set of variables for the more fully adjusted models using quartile cut points for all the sexrace groups and types of CVD diagnostic subsets The 2000 Census data were used to create the population-weighted quartile categories to be used in the statistical analyses The quartile cut points are shown below

population density(persons per square mile)

median 1999 household income

with less than a college education

with Hispanic ethnicity

Hospital distance

Quartile 1 61 - lt507 $2499-lt$40968 0 - lt31 0 - lt2 lt25 (median) Quartile 2 507 - lt102 $40968 - lt54650 31 - lt43 2 - lt5 25 - lt6 (50-75) Quartile 3 102 - lt6410 $54650 - lt72944 43 - lt55 5 - lt10 6 ndash 11 (75-90) Quartile 4 6410 - 94372 $72944 - 200001 55 - 100 10 - 86 11 ndash 45 (90-100)

SAS version 9 provided the statistical programs used to estimate hospitalization rates for exposure proxy categories and demographic categories The models provide ldquocontrast

12

estimatesrdquo ie rate ratios for one category versus a reference category of a variable and 95 confidence intervals (CI) for these rate ratios For the location categories rate ratios for the areas adjacent are compared to areas not adjacent to the river (reference) and areas closest and close to the river are compared to areas distant from the river (reference) If a rate ratio is greater than 100 the model has estimated an elevation of hospitalizations in that category of the study population compared to the reference population If the rate ratio is less than 100 then there are fewer than expected hospitalizations compared to the reference population

The magnitude of the elevation or deficit is also estimated by the rate ratio For instance if the hospitalization rate is twice as high in the adjacent population it would result in a rate ratio of 200 while a 50 higher rate would result in a rate ratio of 150 If the hospitalization rate in the comparison group was only half as big as in the reference group this would result in a rate ratio of 050 The 95 confidence interval assists with determining whether the observed differences are statistically meaningful ie statistically significant The confidence interval provides the range in which there is a 95 probability of including the true rate ratio

In addition to directly assess potential confounding we examined the associations between income and exposure proxy categories and between income and CVD hospitalization risk Because the modeling process provides rate ratios that provide comparisons among groups but not the rates themselves we also calculated the rates per population for each sex race and income category in order to directly observe the relative magnitudes of CVD hospitalization rates Finally analyses stratified by income quartile were also conducted in order to assess the potential role of confounding by income The stratified analyses consist of negative binomial regressions conducted separately for each quartile of income

4 RESULTS

Demographics Associations with Proximity to the River

Study Population Table 4 provides detailed demographic characteristics for the total study population and the populations of the adjacent versus non-adjacent and nearer versus farther from the river block groups from the 2000 Census The total population of the study area for this one year of the study time period is 2924631 persons The first row of Table 4 shows that approximately 11 percent of the study area population is classified as living in block groups adjacent to the Hudson River and 89 live in non-adjacent block groups Approximately 8 of the studyrsquos population lives in the closest block groups approximately 10 lives in close block groups and approximately 82 in distant block groups The table also shows that approximately 81 of the total study population is identified as white 9 black 3 Asian less than 01 Pacific Islander 02 American Indian 4 ldquootherrdquo and 2 ldquomultiple racesrdquo

The 2000 Census data (Table 4) also show that the populations in block groups adjacent and nonshyadjacent to the river are similar in terms of the percent of minority population but the adjacent block groups include a smaller percentage of persons who have attended college and a higher percentage of households below the poverty level Median income is $47320 for block groups adjacent versus $62668 for block groups not adjacent to the river Data for the study area from the 1990 Census show similar distributions (data not shown)

13

For the distance to the river categories there is a gradient for the three levels of distance from the river with the closest block groups on average having higher percentages of minorities smaller percentages of persons who have attended college and higher percentages of households below the poverty level (Table 4) The populations of the close and closest block groups are more similar to each other than to the populations of the distant block groups The population of the closest block groups is estimated to be 19 black close block groups 14 black and distant block groups 8 black The proportions for the ldquootherrdquo category used in the analyses which includes multiple races and American Indians are approximately 12 in both the closest block groups and in the close block groups compared to 5 in the block groups most distant from the river Median income for the closest block groups is $40748 close block groups $44962 and most distant block groups $64847

The bar chart below illustrates the correlations between US Census 2000 indicators of block group socio-economic status and block group centroid (population weighted) distance from the river for the study population

2000 Census Demographics for Study Population by Distance to River Category

black other hispanic no college below poverty

of pop ulati on

0

10

20

30

40

50

60

lt12 mile

12 to 1 mile

gt1 mile

Table 5 shows selected demographic information for the person-year estimates used in the study analyses for the 1990-2005 study period Person-year data represent individuals from Census estimates counted repeatedly for each year of the study timeframe As described in the Methods section the estimation of person years used interpolation to account for population changes from 1990-2000 and techniques for taking account of block boundary changes The person-years estimates for demographic categories (complete data not shown) show similar patterns as the 1990 Census data (data not shown) and 2000 Census data (Table 5) Table 5 for example shows nearly identical percentages of population in each of the river distance categories as Table 4

Table 5 also shows person year information for income quartiles and exposure proxy categories as well as race and exposure proxy categories that is consistent with the Table 4 data for the

14

2000 population Table 5 row percentages show for example that while approximately 11 of the total study population live in adjacent block groups approximately 19 of the population in the lowest income quartile live in adjacent block groups By design approximately 25 of the total study population is in each of the income quartiles Table 5 column percentages show however that in the adjacent block groups 41 of the population is in the lowest income quartile and only 9 are in the highest income quartile

For the block groups within frac12 mile of the river the distribution among income quartiles is even more skewed towards the lowest income quartile 61 of the population living within frac12 mile of the river is in the lowest income quartile and only 6 is in the highest income quartile Given that approximately 25 of the total study population is within each income quartile these percentages indicate that living within the block groups closest to the river is associated with more than a doubling of the likelihood of being in a block group in the lowest socio-economic status quartile 436 of those living in the block groups close to the river (12 to one mile from the river) are in block groups in the lowest income quartile While less than a doubling this indicates that living in the close block groups is associated with a 75 increase in the likelihood of being in the lowest quartile for socio-economic status

Hospitalizations Table 6 shows the process of removing hospitalization records from the study for a variety of reasons The first column includes the 202272 records that contained a residential address and were located within the 12-county study area as described in the Methods section The next five columns show the records remaining after data cleaning exclusion of transfers elimination of records with missing covariates and exclusions due to poor geocoding results Approximately 33000 records were eliminated because they were transfers from one hospital to another for the same event For approximately 1100 records geocoding was inadequate for assigning block group for imputing race as described in the Methods section 336 records were missing sex (11) or age (325) Of the 167733 records remaining 154220 (92) were able to be adequately geocoded to a street address for assignment in a specific block group While approximately eight percent of hospitalizations were eliminated due to inadequate geocoding Table 6 shows that the percentages of hospitalization records for white black Asian and ldquootherrdquo categories and the distribution among diagnostic categories change very little after these records were dropped

The final column of Table 6 shows that of the 154220 hospitalizations in the analysis 101012 IHD hospitalizations comprise the largest health outcome category and are 65 of the hospitalizations in the study AMI hospitalizations (35811) account for 35 of the IHD hospitalizations and angina (4731) accounts for five percent of the IHD hospitalizations Stroke hospitalizations (53208) comprise 34 of the studyrsquos hospitalizations Approximately 36 of the stroke hospitalizations are for ischemic stroke (19371) and approximately 11 are for hemorrhagic stroke (5995)

Table 7 shows the distributions of hospitalizations among the four income quartiles While each income quartile (by definition) contains approximately 25 of the study population for all types of hospitalizations we see that the lowest income quartile includes more than 25 of hospitalizations from a minimum of 286 (IHD) to a maximum of 356 of hospitalizations (angina) The highest income quartile includes fewer than 25 of hospitalizations for each CVD

15

category ranging from 182 for angina to 224 for hemorrhagic stroke These deviations from the 25 quartile distribution show a strong linear pattern with the largest percentage of hospitalizations coming from residents of block groups in the lowest income quartile and the smallest percentages coming from residents of the most wealthy block groups For AMI the lowest income quartilersquos share of hospitalizations is 50 higher than the highest income quartilersquos share of hospitalizations (30 vs 20) IHD and stroke show similar differences and angina shows a doubling with the lowest income quartile contributing 36 of hospitalizations compared to the highest income quartilersquos 18 share of hospitalizations

Table 7 also shows the distribution of hospitalizations among the race categories and the study population race category distribution for comparison Blacks show lower percentages of IHD and AMI than their population percentages and higher percentages of angina and stroke While full presentation of race and sex findings is beyond the scope of this report the lower than expected percentages of IHD and AMI among blacks who on average have lower socioshyeconomic status and are at greater risk for CVD hospitalizations raised questions that were addressed by calculating hospitalization rates per 1000 population for each sex and race group for each income quartile (Appendix B Table 2) Figures 4 5 and 6 show these rates for the white black and Asian categories (These rates for the ldquootherrdquo category are comparatively very high and are not shown on these graphs This issue is addressed in the Discussion section) Figures 4 5 and 6 show the consistent and strong effects of income quartile on hospitalization rates for AMI and stroke for whites blacks and Asians of both sexes The other outcomes show similar patterns (The ldquootherrdquo group showed an opposite pattern [See Discussion Section])

Demographics Summary

The strong associations of lower income with residence closer to the river (Table 5) and lower income with higher hospitalization rates (Tables 7 amp 8) create a difficult challenge for this investigation The multivariable analyses of hospitalization rates attempt to control for socioshyeconomic differences However because these income differences which strongly affect hospitalization rates are also strongly correlated with the gradations of exposure (river distance) it will be difficult to disentangle the influence of socio-economic status from the influence of potential exposures

Statistical Analyses Proximity to the River and CVD Hospitalization Risk

All the variables described above were included in the multivariable regression models and almost every quartile category for every variable showed statistically significant and expected types of effects for all the CVD hospitalization categories (Appendix B Tables 1a and 1b) The following presentation of results focuses on the rate ratios for the exposure proxy variables Table 8a (IHD) and 8b (stroke) show the regression modeling results Results are shown first for models that include only age sex and race in addition to the exposure proxy categories and second for models that also include median income education population density Hispanic ethnicity and distance to nearest hospital The tables provide numbers of hospitalizations and person-years to assist with interpretation For example for total IHD we can see that while 108 of the study population lives in adjacent block groups 122 of the hospitalizations are among people in adjacent block groups This difference of 14 percentage points is the basis for

16

the finding of a statistically significantly elevated rate ratio (risk) for IHD among people living in the adjacent block groups (RR=133 CI 112-157)

Looking across the columns of Tables 8a and 8b we can see the rate ratios for each of the outcomes for the adjacent versus non-adjacent block groups and the results are similar to those for IHD The elevated rate ratios range from 123 for angina to 136 for AMI The only rate ratio not elevated in these first models adjusting only for age sex and race is for hemorrhagic stroke Going down the rows of the Table the next set of rate ratios are for the distance to the river categories The findings are very similar to those for the adjacent block groups Here however we have two exposure proxy categories allowing comparison of the two These rate ratios show no evidence that living in the block groups closest to the river (lt12 mile to the river) produces greater risk than living in the block groups close to the river (12 to 1 mile from the river)

The next set of rate ratios are for models with additional adjustment for the demographics of each hospitalized personrsquos block group as well as distance to the nearest hospital As expected adjustment for block group median income and other demographics accounts for some of the variability in hospitalization counts and thereby reduces the magnitude of the rate ratios estimated for the river adjacency and distance categories The rate ratio elevation for total IHD for example is reduced from 133 to 107 (RR=107 CI 104-111) The elevated rate ratios in these more fully adjusted models for the adjacency analyses are of similarly small magnitude ranging from 106 for AMI to 110 for ischemic stroke Four of the six outcomes show statistically significant elevations Again only hemorrhagic stroke shows no evidence of an elevated rate ratio

For the river distance categories none of the rate ratios show any evidence that living in the block groups closest to the river (lt12 mile to the river) produces greater risk than living in the block groups close to the river (12 to 1 mile) Of the six types of outcomes only two show statistically significantly elevated rate ratios for living in the closest block groups (total stroke and ischemic stroke) For the close block groups four out of six outcomes show statistically significantly elevated rate ratios The statistically significantly elevated rate ratios vary from 106 (AMI stroke) to 113 (angina)

To address the strong evidence described earlier for confounding in these data ie the association of lower income with both CVD hospitalizations and residence closer to the river we also conducted the analyses separately for each quartile of income These results are presented in Table 9 The first results are from analyses adjusting only for age sex and race and the second set of results are from models including the additional demographic factors For simplicity rate ratios are shown only for block groups closest to the river compared as in the other models to block groups farther than one mile from the river (Results were similar for the other distance categories (data not shown)

For the two lowest income quartiles Table 9 shows results similar to those for the study population as a whole (Table 8a 8b) However for the two highest income quartiles which represent frac12 of the study population Table 9 shows a consistent pattern of reduced rate ratios for residence in the closest block groups within frac12 mile of the river In the highest income quartile

17

studies published in 2005-2006 did not provide specific information such as environmental or biomonitoring data in support of the assumption that residence in ZIP codes containing hazardous waste sites with POPs or residence in ZIP codes adjacent to the Hudson River resulted in increased exposures to POPs in the population For the purpose of this follow-up we did not conduct environmental or human biological sampling to address this issue Rather this follow-up replicates the prior studiesrsquo approach with some enhancements to better control for confounding particularly confounding by socio-economic status

This descriptive epidemiologic investigation uses already existing hospitalization data from the Statewide Planning and Research Cooperative System and population data from the United States Census to evaluate comparative rates of cardiovascular hospitalizations While this type of investigation can not show proof of cause and effect it can assist in interpreting relationships among indicators of CVD burden and hypothesized exposures associated with distance from the Hudson River This descriptive investigation may also assist in determining whether additional health investigations of these associations are warranted

2 BACKGROUND

Cardiovascular Disease

The widely known risk factors for cardiovascular disease (CVD) in addition to age sex race and family history include tobacco use high blood pressure high serum cholesterol alcohol use obesity low fruit and vegetable intake physical inactivity diabetes low socio-economic status mental ill-health psychosocial stress and the use of certain medications (WHO 2010 NYS DOH 2009 Wing 1988 NYS DOH 2009a)

CVD Definitions CVD is a general term for any disease of the circulatory system The two main types of CVD are ischemic heart disease also termed coronary heart disease and cerebrovascular disease also called stroke Ischemic heart disease (IHD) results from restricted blood flow through the arteries supplying the heart muscle The most serious risk from IHD is acute myocardial infarction (AMI) another term for heart attack AMI occurs when blood flow to the heart is suddenly reduced or stopped usually due to a blockage in a coronary artery Angina is severe constricting pain in the chest often radiating from the area immediately over the heart to the left shoulder and down the arm Cerebrovascular disease refers primarily to stroke the sudden impairment of brain function resulting from interruption of circulation to the brain Ischemic stroke results from blockage of an artery supplying blood to the brain Hemorrhagic stroke occurs due to the escape of blood from a ruptured artery supplying blood to the brain

CVD Burden and Trends CVD is the leading cause of death for men and women of all ages in the United States and in New York State In 1999 for every death from IHD there were five hospitalizations for ischemic disease in NYS (NYS DOH 2009a) Mortality rates from CVD have been declining steadily since the 1980rsquos Research studies estimate that approximately half of the decline is due to improved treatments for CVD and its risk factors such as high cholesterol and high blood pressure and approximately half is due to decreasing incidence of CVD The decrease in incidence is due to trends in health behaviors primarily reduced smoking rates

3

CVD mortality rates do not appear to be declining for all segments of the population however and may actually be increasing among some groups particularly people of lower socio-economic status (Barnett et al 1999) This may be due in part to higher rates of smoking among people of lower socio-economic status as well as to other factors related to socio-economic status and race A growing literature points to multiple causes for CVD and other health disparities including health practices psychosocial stress limited resources discrimination and access to health care (Health United States 2007) Increases in the prevalence of diabetes and obesity in the general population partially offset advances in treatment and favorable trends for other risk factors An increasing ischemic heart disease mortality rate in women ages 35 to 54 since 1997 is attributed to these unfavorable risk factor trends (Ford 2007 Lloyd-Jones 2009) New York State (NYS) and other northeast states have high IHD and low stroke mortality with NYS having the highest IHD mortality and lowest stroke mortality among the 50 states (Howard et al 2009)

CVD Data Sources CVD mortality rates are more easily measured than incidence prevalence or hospitalizations because mortality data from vital records are a comprehensive and relatively accessible data source Information for estimating the incidence and prevalence of CVD outcomes comes primarily from health interview physical examination and record surveys Hospitalization data are available for NYS and the US but are generally collected and analyzed in order to track trends in hospital care and resources expended Actual CVD incidence data validated by physician diagnosis physician examination or reviews of medical records are not directly available for the entire US or for specific geographic areas of NYS Such information is available from population-based cohort studies that follow people over time in selected communities While the concern underlying this research project is to understand CVD levels in the community ie incidence and prevalence we do not have data that directly provide us with this information In the absence of such data hospitalizations are frequently used as indicators of the burden of disease in communities particularly in surveillance or hypothesis-generating investigations such as this one In NYS the Behavioral Risk Factor Surveillance System (BRFSS) an ongoing monthly telephone survey of adults aged 18 years and older provides information about the burden of CVD by age sex race income and education

CVD Risk Factor Data The BRFSS data for NYS show that over 20 of the population aged 65 and older reported having been diagnosed with CVD versus less than five percent of the population under 65 Males were slightly more likely to report having CVD than females Nonshyhispanic whites compared to other ethnic groups combined were slightly more likely to report having CVD None of these slight differences were statistically significant however The only statistically significant demographic risk factors in the BRFSS data were low income and education levels Statistically significant elevations of reported CVD were observed for respondents with incomes below $25000 and respondents with a high school degree or less education (NYS DOH 2009a)

Another factor that can affect the outcome measure hospitalizations is the likelihood of mortality prior to reaching the hospital Studies show a variety of factors affect whether an individual experiencing a heart attack or stroke outside the hospital survives to be counted in hospitalization data These factors include lack of knowledge about personal risk or prior disease socio-economic status race marital status and distance to the hospital (Galea 2007

4

Barnett 2006 Ayala 2003) Population-based studies report sudden cardiac death usually outside the hospital to be the first overt manifestation of heart disease for 40 to 60 of all cases ( Kannel 1987 deVreede-Swagemakers 1997 UK Heart Attack Study 1998) An analysis of out-of-hospital mortality from AMI in Pennsylvania in 1998 showed that in relatively rural areas where travel time to the hospital was 25 minutes or longer 72 of all AMI deaths occurred outside the hospital compared with a statewide average of 49 of AMI deaths occurring outside of the hospital (OrsquoNeill 2003) In New York State 55 of IHD deaths occur outside the hospital compared to 35 of stroke deaths (NYS DOH 2009b)

Hudson River PCB Contamination

During an approximate 30-year period ending in 1977 the General Electric (GE) plants in Hudson Falls and Fort Edward used PCBs in the manufacture of electrical capacitors and discharged large quantities of PCBs into the upper Hudson River Estimates of the total direct discharges to the river are as high as 1330000 pounds In 1983 the US EPA classified a 200shymile section of the Hudson River from Hudson Falls to New York City as a National Priority List site making it one of the largest Superfund sites in the United States In 2002 the US EPA issued a Record of Decision calling for targeted environmental dredging and removal of PCB-contaminated sediment in the upper River (USEPA 2002) The dredging began in May 2009

The Superfund environmental investigations and planned remedial actions have focused on the areas of the river near Hudson Falls and Fort Edward where the highest levels of PCBs in sediments are located The 2002 Record of Decision provides information about the human health risk assessment conducted for the Upper and Mid-Hudson and the ecological risk assessment conducted for the Upper and Lower Hudson River (USEPA 2002) Fish contamination and fish consumption are the primary concerns at the site

In 1976 fishing was banned in the upper river and in 1995 a catch and release policy replaced the ban NYS DOH continues to recommend people eat no fish from the Upper Hudson River that children under age 15 and women of childbearing age eat no fish from the entire 200 mile stretch of river below Hudson Falls and that the general population eat none of most species of fish caught between the federal dam at Troy and Catskill approximately 40 miles south of Troy The Lower Hudson remains closed to commercial fishing for striped bass and eight other species

The fishing bans and advisories were issued because ingestion of fish is the major potential pathway for exposure for the general population Many previous studies have shown relationships between consumption of fish from contaminated waters and human PCB levels Another potential pathway for exposure less frequently studied to date is inhalation of PCBs NYS DOH and ATSDR launched ldquoThe Hudson River Communities Projectrdquo in 2000 to address this issue (More information on this project is provided below)

POPS and PCBs

The category ldquopersistent organic pollutantsrdquo (POPs) includes man-made chemicals that are by definition persistent in the environment Examples of POPs are polychlorinated biphenyls

5

(PCBs) polychlorinated dibenzo-p-dioxins (PCDDs) polychorinated dibenso-furans (PCDFs) and pprsquo-diphenyldichloroethene (DDE) a breakdown product of the pesticide dichlorodiphenyltrichloroethane (DDT) These chemicals are stored in human fat tissues accumulating over an individualrsquos lifetime They also accumulate in fish and other species thereby concentrating up the food chain Levels of most POPs measured in blood serum of Americans have decreased in the last few decades An exception is polybrominated diphenyl ethers (PBDEs) levels of which have been increasing in recent years While levels of many POPs are declining a large number continue to be detectable in blood for most of the US population (USDHHS 2009)

This follow-up investigation focuses on the Hudson River where the POPs of concern are PCBs PCBs are a group of synthetic chlorinated compounds manufactured in the US through 1977 Production was halted due to concerns about the persistence of PCBs in the environment and the toxicity of PCBs shown in animal models PCBs were used primarily by the electrical industry for capacitors and transformers and were also used in hydraulic fluids fluorescent light fixtures flame retardants inks adhesives carbonless copy paper paints pesticide extenders plasticizers and wire insulators PCBs can be oily liquids or solids and some are volatile and can exist as vapor in air PCBs in the environment are generally mixtures of the many different types known as congeners (ATSDR 2000)

The variety of types of PCBs adds to the complexity of studying adverse animal and human health effects Increasingly PCB research focuses on specific PCB congeners and groups of congeners because specific types of congeners appear to have different mechanisms of action and toxicity Dioxin-like PCBs (DLPCBs) are an important subset of PCBs due to concerns about their toxicity The focus on dioxin-like PCBs is paralleled by increasing research on the group of POPs with dioxin-like properties This type of PCB and POP research assigns toxic potency values relative to dioxin to various PCBs and POPs to provide an overall toxic equivalency value to specific PCB and other POPs

From 2002 through 2007 NYS DOH and SUNY Albany School of Public Health researchers conducted the Hudson River Communities Study among residents of Hudson Falls and Fort Edward the villages where the PCB discharges from two GE electrical capacitor plants occurred The comparison community was upriver from the discharges in Glens Falls This studyrsquos evaluation of exposures included measuring PCBs in outdoor air and in blood serum The study found that PCB levels in outdoor air were statistically significantly higher in the Hudson FallsFort Edward area (072 nanograms per cubic meter) than in the upriver Glens Falls area (040 nanograms per cubic meter) but that the PCB concentrations in the Hudson FallsFort Edward area were lower than those in other communities with known PCB-contaminated sites and similar to levels reported in other locations in the northeastern United States A statistically significant reduction in the total PCB concentrations in outdoor air samples was observed at a distance of 1200 meters (34 mile) from the river from 0760 ngm3 to 0497 ngm3 (Fitzgerald et al 2007 Palmer et al 2008)

The average PCB concentrations in serum for the study and comparison populations did not differ significantly (307 ppb wet weight versus 323 ppb) The average toxic equivalency quantities (TEQ) for dioxin-like PCB congeners for the study and comparison area populations

6

also did not differ In addition the study found no detectable differences in study subjectsrsquo serum PCB levels associated with proximity to contaminated portions of the river or other PCB sources The study did find that lifetime consumption of Hudson River fish was associated with higher serum PCB levels (Fitzgerald 2007 Palmer 2008) Another recent biomonitoring study of Brooklyn Manhattan and New Jersey fisherman who consumed versus those who did not consume their catch from the lower Hudson River and New YorkNew Jersey harbor suggests eating fish from these waters is not associated with an increased body burden of PCBs or most other organochlorines This study did not address the inhalation pathway (Morland 2008)

PCBs and Cardiovascular Disease Outcomes

Research studies on adverse health effects associated with PCB exposure have focused on many types of outcomes organs and biological systems including cancer liver reproductive and developmental effects thyroid endocrine immune system and neurological effects as well as specific effects on the skin and eyes Relatively little research has focused to date on associations between PCB exposures and CVD outcomes The majority of human studies have assessed health effects in occupational groups having relatively high-level exposures among people consuming fish from contaminated waters or people exposed to relatively high levels due to accidental contamination of food (USDHHS NIH 2010)

PCBs have been shown to produce a variety of toxic effects in animals These effects include cancer allergies hypersensitivity damage to the central and peripheral nervous systems reproductive disorders and disruption of the immune system (ATSDR 2000) In contrast research to date does not tell us whether specific types of human health effects have occurred from the low-level exposures experienced by the general population Some human epidemiological studies of relatively low levels of PCDDs PCDFs and PCBs suggest possible relationships not only with neurodevelopment of infants but also diabetes and a condition called insulin resistance conditions that increase CVD risk (Longnecker 2001 Carpenter 2006 Arisawa 2005)

In 1998 a study of a population in Seveso Italy exposed to relatively pure TCDD (dioxin) from an industrial accident showed increased mortality from CVD respiratory diseases and diabetes (Bertazzi 1998) These findings suggested the possibility that other dioxin-like chemicals such as dioxin-like PCBs might increase risk for these diseases Some more recent animal studies described below suggest possible mechanisms of action for PCB effects that increase risk for CVD diabetes and hypertension Another relevant source of information is from research using the biological monitoring data from the National Health and Nutrition Examination Survey (NHANES) project to look at the association between PCB levels in blood and self-reported diabetes and heart disease The body of evidence based on analytical human epidemiological studies remains relatively sparse however

Some recent studies using NHANES biomonitoring data show associations between PCB levels and CVD (Lee 2006 Lee 2007 Ha 2007 Everett 2008) These studies are cross-sectional (comparisons at one point in time) evaluations of PCB levels measured in blood samples from individuals and self-reported CVD and diabetes a risk factor for CVD While these studies use individual-level biomonitoring and self-reported health outcome data they are considered to be

7

relatively weak in terms of providing evidence of causality because both exposure and outcome are measured at the same point in time Given the nature of NHANES data collection and reporting practices these studies are not able to control for possible geographical confounding which may occur if urban versus rural residence is associated with higher PCB levels and higher risk for CVD for example The higher PCB levels among African Americans in these data suggest urban residence may be playing a role in the PCB data

The results of animal studies suggest multiple mechanisms by which PCB exposures could affect cardiovascular risk as well as multiple types of adverse health outcomes One hypothesized mechanisms of action specifically for dioxin-like PCBs is that they may have anti-estrogenic effects which impact CVD risk factors such as serum cholesterol levels and blood pressure (Lind 2004) Several studies suggest PCBs interact with a specific receptor (AhR) and increase cellular oxidative stress causing inflammation cell injury and cardiovascular dysfunction (Hennig 2002) Similarly some studies suggest PCB-induced inflammation of fat cells leads to obesity and increased risk for coronary artery disease (atherosclerosis) (Arsenescu 2008)

3 FOLLOW-UP STUDY DATA SOURCES AND METHODS

This follow-up investigation will provide perspective on the 2004-2005 findings (Sergeev and Carpenter 2005 Shcherbatykh et al 2005) by producing results from analyses conducted at a more detailed level of geography the block group rather than ZIP code in order to better control for confounding When the term confounding is used by epidemioogists it refers to the problem that can arise when there is a factor such as income level that is not the factor of interest in the study but that is related to the risk factor being studied (residential location ndash proxy for exposure) and to the disease being studied (CVD) This type of factor is labeled a ldquoconfounderrdquo because it can confound ie interfere with or confuse the evaluation of associations among the variables of interest in the data All types of epidemiological studies face issues of confounding but ecological ie group-level analyses face additional challenges related to potential confounding because analyses can not account for some important individual differences within the groups that are the units of analysis (Morganstern 1998b Hertz-Picciotto 1998) This investigation was able to control for each individualrsquos age sex and race but not for individual-level risk factors such as income or education level

The follow-up investigationrsquos specific objectives are to

improve the ability to control for confounding by socio-economic status through the use of block groups rather than ZIP codes as the unit of analysis

compare the hospitalization rates among residents of block groups nearest the Hudson River to those among residents of block groups farther from the Hudson River

compare the findings of this follow-up study to those of the earlier studies

Study Area and Exposure Areas

The overall study area is comprised of the 12 counties with geographic areas abutting the Hudson River south of the source of PCB contamination from the General Electric facility in Hudson Falls The 12 counties are Westchester Rockland Putnam Orange Dutchess Ulster

8

Columbia Greene Rensselaer Albany Washington and Saratoga US Census boundaries are used to define the study area because census population estimates provide the data for the study area populationrsquos size and the populationrsquos age sex and race distribution Because the study period spanned 1990 through 2005 and populations change over time 1990 and 2000 Census data were used to linearly interpolate the between-year populations from 1991 through 1999 for each block group 2000 data were used without extrapolation to estimate block group populations from 2001 through 2005

Because block group boundaries change between the Census years 1990 block populations were normalized to fit 2000 block group boundaries before the estimation of sex age and race group specific populations In the digital map 2000 block group boundaries were overlaid on the 1990 block centroids to assign 1990 block population counts to the corresponding 2000 block group boundaries Then 1990 block populations were summed to fit the block group boundaries Next population counts by age group (0-14 15-24 25-34 35-44 45-54 55-64 65-74 75+) sex (male female) and race (white black asian other) were estimated for each year at the block group level This investigation includes all ages over 25 years and the study population is divided into four race categories white black or African-American Asian and ldquoOtherrdquo The Pacific Island group was combined with the Asian group and American Indian ldquootherrdquo and ldquomultiple racesrdquo categories are combined into a larger ldquootherrdquo category

Classification of the population into potential exposure categories for this study is based exclusively on location relative to the Hudson River Two types of exposure proxy categories were developed block group adjacency and block group centroid distance to the river shore For the adjacency categories block groups were assigned to the adjacent category if they directly abutted the Hudson River and all other block groups in the 12 counties were assigned to the nonshyadjacent category

For the distance categories a central point (centroid) within each block group was identified and used to assign block group centroid distance to the river shore The block group centroids were population-weighted which means the geographic distribution of households within 2000 census blocks were used to locate block group centroids based on the geographic population distribution By locating a population-weighted block centroid rather than a more straightforward distance centroid the assignment of block group distance to the river shore more accurately reflects distance to the river for the majority of block group residents Buffers of one-half mile and one mile from the Hudson River boundary were created for the assignment of each block group into one of the three distance categories less than frac12 mile (closest) frac12 to one mile (close) and farther than one mile from the river (distant)

Distance categories were selected a priori at the start of the study Because environmental sampling data were not collected along the length of the river included in this investigation we are not able to determine whether these distance categories represent actual differences in exposures As discussed previously a statistically significant reduction in the total PCB concentrations in outdoor air samples was observed at a distance of 1200 meters (34 mile) from the river from 0760 ngm3 to 0497 ngm3 in a study conducted in the Hudson Falls Fort Edward area near the source of PCBs (Fitzgerald 2007 Palmer 2007)

9

For the adjacency categories of the 2371 block groups in the study area 255 (11) were classified as adjacent to the Hudson River and 2116 (89) were classified as non-adjacent (Figure 2) For the distance categories 225 (9) were classified as closest to the river 216 (9) as close and 1930 block groups (81) as distant from the river

Comparison of Figures 1 and 2 show that block groups include smaller geographic areas than ZIP codes and that block groups are relatively smaller in area in urban versus rural areas The study area includes a total of 368 ZIP codes compared to a total of 2371 block groups ZIP code average population in the study area is approximately 8000 compared to block group average population of 1200 Comparison of Figures 2 and 3 shows that using population-weighted centroid distance to the river produces different exposure areas than adjacency Fifty-three (20) of the adjacent block groups were categorized for the distance indicator as distant because the population in the block group was concentrated in the area farther than one mile from the river On the other hand about 11 of the 2116 non-adjacent block groups were classified as closest (77) or close (162)

Hospitalization Data

Hospitalizations for ischemic heart disease (IHD) and cerebrovascular disease (stroke) were included in this study More specifically hospital discharge records that listed as the principal diagnosis either IHD (ICD-9-CM 410-414 and 4297) or cerebrovascular disease (ICD-9-CM 430-436) and subsets of each of these categories similar to the subsets evaluated in the Sergeev et al and Shcherbatykh studies were included The subsets are AMI and angina subsets of IHD and ischemic and hemorrhagic stroke which are subsets of cerebrovascular disease (stroke) (Principal diagnosis represents the illness for which the person was admitted to the hospital) Angina is an outcome not evaluated in the prior studies It was added here for comparative purposes because it represents an outcome appropriately addressed by primary outpatient rather than inpatient hospital care Angina is one of a group of outcomes included in the recently developed national ldquoPrevention Quality Indicatorsrdquo (USDHHS AHRQ 2007) as an indicator for geographic areas or populations for whom primary care is inadequate Table 3 lists the specific ICD-9-CM codes and definitions included in this investigation

While one of the previously published studies (Huang et al 2006) showed elevations in hypertension (high blood pressure) rates this outcome is rarely listed as the principal reason for hospitalization As a secondary diagnosis however hypertension is extremely common with substantial overlap with the primary diagnoses of IHD and stroke In addition hypertension hospitalizations were not as elevated in the ZIP codes adjacent to the Hudson River as the IHD and stroke outcomes As a practical matter geocoding the hypertension hospitalization data (1364000 hospitalizations) would overwhelm the resources for this project For these reasons hypertension was not included in this follow-up investigation

The source of the hospitalization data was the NYS DOH Statewide Planning and Research Cooperative System (SPARCS) established in 1979 to collect detailed records on discharges from hospitals located in New York State Persons seen in the emergency department but not admitted are not included in the SPARCS data Hospitalizations at US Veterans Administration

10

Hospitals are also not included in this dataset We identified the hospital transfers in the data and excluded them so that individuals are only counted once for the same event

The NYS DOH Data Protection Review Board for SPARCS hospitalization data approved our use of hospitalization records for this investigation and provided approximately 460000 records with a principal diagnosis of either IHD or stroke Because of the large number of cardiovascular hospitalizations and this studyrsquos requirement for geocoding each hospitalized personrsquos address we selected a 50 random sample of the hospital discharges with principal diagnoses of IHD or stroke for inclusion in the study (229948 records) Approximately one percent of these records (2737) were eliminated from the study because they contained no residential address resulting in 227211 hospitalization records 24939 of these records were eliminated after geocoding because they were not within the 12 county study area This left 202272 records remaining in the study

For geocoding to the block group level the hospitalization street addresses were modified as needed using US Postal Service standards The addresses were assigned geographic coordinates using commercially available geocoding software (MapMarker 2004) A combination of land parcel data and street centerline files were used to assign geographic coordinates No contact was made with cases parents legal guardians or next of kin of cases to determine residential locations

Once geographic coordinates were assigned to cases through address-matching the case locations were overlaid onto digital maps of the study area using a geographic information system so that the number of observed cases falling within the study area boundaries could be determined and a 2000 Census block group could be assigned to each hospitalization In order to protect confidentiality no maps of individual case locations are provided Geographic coordinates were not able to be assigned for some addresses for example rural routes or post office boxes Internet searches were conducted for addresses that were located in facilities such as nursing homes and trailer parks

While fewer than one percent of records were missing sex or age information approximately 15000 records seven percent were missing race information Evaluation of the geographic and temporal distribution of these records showed they were not randomly geographically distributed 47 of the records with no race information were from one hospital for that hospital race was missing from 100 of the records in all years except one after 1997 mid-way through the study time-frame Evaluation of other records missing race showed similar patterns of 100 of records missing race in specific years In order to prevent biased hospitalization rate estimates records missing race were assigned race randomly assuming a race distribution equal to that of the individualrsquos residential block group

Statistical Analysis

This investigation used negative binomial multivariable regression models the statistical methods used in the previous studies (Sergeev and Carpenter 2005 [IHD] Shcherbatykh et al 2005 [stroke]) The follow-up studyrsquos analyses compared hospitalization rates for populations in block groups adjacent or in close proximity to the Hudson River to rates for populations more

11

distant from the Hudson River while adjusting for the potential influence of other factors Sex age and race-specific hospitalization count data comprise the dependent variable and the statistical methods attempt to predict the counts using the sex age and race distributions from Census data as well as block group-level demographics such as median income Models were first estimated using basic Census information for age sex and race Then more fully adjusted models were estimated using information at the block group level for income education population density Hispanic ethnicity and distance to the nearest hospital

More specifically each hospitalized personrsquos block group was used to assign quartile categories of block group median household income of population with less than a college education population density of the population identified as Hispanic and distance from the block group centroid to the nearest hospital These variables were selected in advance The population density variable was selected to account for potential effects on hospitalization rates of large differences in population density from the relatively rural northern counties to the more urban southern counties in the study area of the population identified as Hispanic was included because Hispanic ethnicity may be associated with CVD risk but Hispanic ethnicity was not available in the hospitalization records Distance to the hospital was included because studies show that shorter distances increase the likelihood that individuals with AMI and stroke survive to be admitted to the hospital

As described previously the literature about risk factors for CVD shows strong associations between lower income and education and higher CVD risk Regarding the effects of income and education on hospitalization rates the literature from analytic studies specifically of hospitalizations for CVD is quite sparse Most analytic studies focus on prevalence incidence or mortality in population-based studies or cohort studies and there are not many of these An example of a study that was useful as background is a study of income level and CVD mortality that showed strong associations of lower income and higher CVD mortality for men and women with the relationship for men being linear throughout the range of income levels while for women the relationship was only strong at levels of income below the median (Rehkopf 2008)

This follow-up studyrsquos approach was to use the same set of variables for the more fully adjusted models using quartile cut points for all the sexrace groups and types of CVD diagnostic subsets The 2000 Census data were used to create the population-weighted quartile categories to be used in the statistical analyses The quartile cut points are shown below

population density(persons per square mile)

median 1999 household income

with less than a college education

with Hispanic ethnicity

Hospital distance

Quartile 1 61 - lt507 $2499-lt$40968 0 - lt31 0 - lt2 lt25 (median) Quartile 2 507 - lt102 $40968 - lt54650 31 - lt43 2 - lt5 25 - lt6 (50-75) Quartile 3 102 - lt6410 $54650 - lt72944 43 - lt55 5 - lt10 6 ndash 11 (75-90) Quartile 4 6410 - 94372 $72944 - 200001 55 - 100 10 - 86 11 ndash 45 (90-100)

SAS version 9 provided the statistical programs used to estimate hospitalization rates for exposure proxy categories and demographic categories The models provide ldquocontrast

12

estimatesrdquo ie rate ratios for one category versus a reference category of a variable and 95 confidence intervals (CI) for these rate ratios For the location categories rate ratios for the areas adjacent are compared to areas not adjacent to the river (reference) and areas closest and close to the river are compared to areas distant from the river (reference) If a rate ratio is greater than 100 the model has estimated an elevation of hospitalizations in that category of the study population compared to the reference population If the rate ratio is less than 100 then there are fewer than expected hospitalizations compared to the reference population

The magnitude of the elevation or deficit is also estimated by the rate ratio For instance if the hospitalization rate is twice as high in the adjacent population it would result in a rate ratio of 200 while a 50 higher rate would result in a rate ratio of 150 If the hospitalization rate in the comparison group was only half as big as in the reference group this would result in a rate ratio of 050 The 95 confidence interval assists with determining whether the observed differences are statistically meaningful ie statistically significant The confidence interval provides the range in which there is a 95 probability of including the true rate ratio

In addition to directly assess potential confounding we examined the associations between income and exposure proxy categories and between income and CVD hospitalization risk Because the modeling process provides rate ratios that provide comparisons among groups but not the rates themselves we also calculated the rates per population for each sex race and income category in order to directly observe the relative magnitudes of CVD hospitalization rates Finally analyses stratified by income quartile were also conducted in order to assess the potential role of confounding by income The stratified analyses consist of negative binomial regressions conducted separately for each quartile of income

4 RESULTS

Demographics Associations with Proximity to the River

Study Population Table 4 provides detailed demographic characteristics for the total study population and the populations of the adjacent versus non-adjacent and nearer versus farther from the river block groups from the 2000 Census The total population of the study area for this one year of the study time period is 2924631 persons The first row of Table 4 shows that approximately 11 percent of the study area population is classified as living in block groups adjacent to the Hudson River and 89 live in non-adjacent block groups Approximately 8 of the studyrsquos population lives in the closest block groups approximately 10 lives in close block groups and approximately 82 in distant block groups The table also shows that approximately 81 of the total study population is identified as white 9 black 3 Asian less than 01 Pacific Islander 02 American Indian 4 ldquootherrdquo and 2 ldquomultiple racesrdquo

The 2000 Census data (Table 4) also show that the populations in block groups adjacent and nonshyadjacent to the river are similar in terms of the percent of minority population but the adjacent block groups include a smaller percentage of persons who have attended college and a higher percentage of households below the poverty level Median income is $47320 for block groups adjacent versus $62668 for block groups not adjacent to the river Data for the study area from the 1990 Census show similar distributions (data not shown)

13

For the distance to the river categories there is a gradient for the three levels of distance from the river with the closest block groups on average having higher percentages of minorities smaller percentages of persons who have attended college and higher percentages of households below the poverty level (Table 4) The populations of the close and closest block groups are more similar to each other than to the populations of the distant block groups The population of the closest block groups is estimated to be 19 black close block groups 14 black and distant block groups 8 black The proportions for the ldquootherrdquo category used in the analyses which includes multiple races and American Indians are approximately 12 in both the closest block groups and in the close block groups compared to 5 in the block groups most distant from the river Median income for the closest block groups is $40748 close block groups $44962 and most distant block groups $64847

The bar chart below illustrates the correlations between US Census 2000 indicators of block group socio-economic status and block group centroid (population weighted) distance from the river for the study population

2000 Census Demographics for Study Population by Distance to River Category

black other hispanic no college below poverty

of pop ulati on

0

10

20

30

40

50

60

lt12 mile

12 to 1 mile

gt1 mile

Table 5 shows selected demographic information for the person-year estimates used in the study analyses for the 1990-2005 study period Person-year data represent individuals from Census estimates counted repeatedly for each year of the study timeframe As described in the Methods section the estimation of person years used interpolation to account for population changes from 1990-2000 and techniques for taking account of block boundary changes The person-years estimates for demographic categories (complete data not shown) show similar patterns as the 1990 Census data (data not shown) and 2000 Census data (Table 5) Table 5 for example shows nearly identical percentages of population in each of the river distance categories as Table 4

Table 5 also shows person year information for income quartiles and exposure proxy categories as well as race and exposure proxy categories that is consistent with the Table 4 data for the

14

2000 population Table 5 row percentages show for example that while approximately 11 of the total study population live in adjacent block groups approximately 19 of the population in the lowest income quartile live in adjacent block groups By design approximately 25 of the total study population is in each of the income quartiles Table 5 column percentages show however that in the adjacent block groups 41 of the population is in the lowest income quartile and only 9 are in the highest income quartile

For the block groups within frac12 mile of the river the distribution among income quartiles is even more skewed towards the lowest income quartile 61 of the population living within frac12 mile of the river is in the lowest income quartile and only 6 is in the highest income quartile Given that approximately 25 of the total study population is within each income quartile these percentages indicate that living within the block groups closest to the river is associated with more than a doubling of the likelihood of being in a block group in the lowest socio-economic status quartile 436 of those living in the block groups close to the river (12 to one mile from the river) are in block groups in the lowest income quartile While less than a doubling this indicates that living in the close block groups is associated with a 75 increase in the likelihood of being in the lowest quartile for socio-economic status

Hospitalizations Table 6 shows the process of removing hospitalization records from the study for a variety of reasons The first column includes the 202272 records that contained a residential address and were located within the 12-county study area as described in the Methods section The next five columns show the records remaining after data cleaning exclusion of transfers elimination of records with missing covariates and exclusions due to poor geocoding results Approximately 33000 records were eliminated because they were transfers from one hospital to another for the same event For approximately 1100 records geocoding was inadequate for assigning block group for imputing race as described in the Methods section 336 records were missing sex (11) or age (325) Of the 167733 records remaining 154220 (92) were able to be adequately geocoded to a street address for assignment in a specific block group While approximately eight percent of hospitalizations were eliminated due to inadequate geocoding Table 6 shows that the percentages of hospitalization records for white black Asian and ldquootherrdquo categories and the distribution among diagnostic categories change very little after these records were dropped

The final column of Table 6 shows that of the 154220 hospitalizations in the analysis 101012 IHD hospitalizations comprise the largest health outcome category and are 65 of the hospitalizations in the study AMI hospitalizations (35811) account for 35 of the IHD hospitalizations and angina (4731) accounts for five percent of the IHD hospitalizations Stroke hospitalizations (53208) comprise 34 of the studyrsquos hospitalizations Approximately 36 of the stroke hospitalizations are for ischemic stroke (19371) and approximately 11 are for hemorrhagic stroke (5995)

Table 7 shows the distributions of hospitalizations among the four income quartiles While each income quartile (by definition) contains approximately 25 of the study population for all types of hospitalizations we see that the lowest income quartile includes more than 25 of hospitalizations from a minimum of 286 (IHD) to a maximum of 356 of hospitalizations (angina) The highest income quartile includes fewer than 25 of hospitalizations for each CVD

15

category ranging from 182 for angina to 224 for hemorrhagic stroke These deviations from the 25 quartile distribution show a strong linear pattern with the largest percentage of hospitalizations coming from residents of block groups in the lowest income quartile and the smallest percentages coming from residents of the most wealthy block groups For AMI the lowest income quartilersquos share of hospitalizations is 50 higher than the highest income quartilersquos share of hospitalizations (30 vs 20) IHD and stroke show similar differences and angina shows a doubling with the lowest income quartile contributing 36 of hospitalizations compared to the highest income quartilersquos 18 share of hospitalizations

Table 7 also shows the distribution of hospitalizations among the race categories and the study population race category distribution for comparison Blacks show lower percentages of IHD and AMI than their population percentages and higher percentages of angina and stroke While full presentation of race and sex findings is beyond the scope of this report the lower than expected percentages of IHD and AMI among blacks who on average have lower socioshyeconomic status and are at greater risk for CVD hospitalizations raised questions that were addressed by calculating hospitalization rates per 1000 population for each sex and race group for each income quartile (Appendix B Table 2) Figures 4 5 and 6 show these rates for the white black and Asian categories (These rates for the ldquootherrdquo category are comparatively very high and are not shown on these graphs This issue is addressed in the Discussion section) Figures 4 5 and 6 show the consistent and strong effects of income quartile on hospitalization rates for AMI and stroke for whites blacks and Asians of both sexes The other outcomes show similar patterns (The ldquootherrdquo group showed an opposite pattern [See Discussion Section])

Demographics Summary

The strong associations of lower income with residence closer to the river (Table 5) and lower income with higher hospitalization rates (Tables 7 amp 8) create a difficult challenge for this investigation The multivariable analyses of hospitalization rates attempt to control for socioshyeconomic differences However because these income differences which strongly affect hospitalization rates are also strongly correlated with the gradations of exposure (river distance) it will be difficult to disentangle the influence of socio-economic status from the influence of potential exposures

Statistical Analyses Proximity to the River and CVD Hospitalization Risk

All the variables described above were included in the multivariable regression models and almost every quartile category for every variable showed statistically significant and expected types of effects for all the CVD hospitalization categories (Appendix B Tables 1a and 1b) The following presentation of results focuses on the rate ratios for the exposure proxy variables Table 8a (IHD) and 8b (stroke) show the regression modeling results Results are shown first for models that include only age sex and race in addition to the exposure proxy categories and second for models that also include median income education population density Hispanic ethnicity and distance to nearest hospital The tables provide numbers of hospitalizations and person-years to assist with interpretation For example for total IHD we can see that while 108 of the study population lives in adjacent block groups 122 of the hospitalizations are among people in adjacent block groups This difference of 14 percentage points is the basis for

16

the finding of a statistically significantly elevated rate ratio (risk) for IHD among people living in the adjacent block groups (RR=133 CI 112-157)

Looking across the columns of Tables 8a and 8b we can see the rate ratios for each of the outcomes for the adjacent versus non-adjacent block groups and the results are similar to those for IHD The elevated rate ratios range from 123 for angina to 136 for AMI The only rate ratio not elevated in these first models adjusting only for age sex and race is for hemorrhagic stroke Going down the rows of the Table the next set of rate ratios are for the distance to the river categories The findings are very similar to those for the adjacent block groups Here however we have two exposure proxy categories allowing comparison of the two These rate ratios show no evidence that living in the block groups closest to the river (lt12 mile to the river) produces greater risk than living in the block groups close to the river (12 to 1 mile from the river)

The next set of rate ratios are for models with additional adjustment for the demographics of each hospitalized personrsquos block group as well as distance to the nearest hospital As expected adjustment for block group median income and other demographics accounts for some of the variability in hospitalization counts and thereby reduces the magnitude of the rate ratios estimated for the river adjacency and distance categories The rate ratio elevation for total IHD for example is reduced from 133 to 107 (RR=107 CI 104-111) The elevated rate ratios in these more fully adjusted models for the adjacency analyses are of similarly small magnitude ranging from 106 for AMI to 110 for ischemic stroke Four of the six outcomes show statistically significant elevations Again only hemorrhagic stroke shows no evidence of an elevated rate ratio

For the river distance categories none of the rate ratios show any evidence that living in the block groups closest to the river (lt12 mile to the river) produces greater risk than living in the block groups close to the river (12 to 1 mile) Of the six types of outcomes only two show statistically significantly elevated rate ratios for living in the closest block groups (total stroke and ischemic stroke) For the close block groups four out of six outcomes show statistically significantly elevated rate ratios The statistically significantly elevated rate ratios vary from 106 (AMI stroke) to 113 (angina)

To address the strong evidence described earlier for confounding in these data ie the association of lower income with both CVD hospitalizations and residence closer to the river we also conducted the analyses separately for each quartile of income These results are presented in Table 9 The first results are from analyses adjusting only for age sex and race and the second set of results are from models including the additional demographic factors For simplicity rate ratios are shown only for block groups closest to the river compared as in the other models to block groups farther than one mile from the river (Results were similar for the other distance categories (data not shown)

For the two lowest income quartiles Table 9 shows results similar to those for the study population as a whole (Table 8a 8b) However for the two highest income quartiles which represent frac12 of the study population Table 9 shows a consistent pattern of reduced rate ratios for residence in the closest block groups within frac12 mile of the river In the highest income quartile

17

CVD mortality rates do not appear to be declining for all segments of the population however and may actually be increasing among some groups particularly people of lower socio-economic status (Barnett et al 1999) This may be due in part to higher rates of smoking among people of lower socio-economic status as well as to other factors related to socio-economic status and race A growing literature points to multiple causes for CVD and other health disparities including health practices psychosocial stress limited resources discrimination and access to health care (Health United States 2007) Increases in the prevalence of diabetes and obesity in the general population partially offset advances in treatment and favorable trends for other risk factors An increasing ischemic heart disease mortality rate in women ages 35 to 54 since 1997 is attributed to these unfavorable risk factor trends (Ford 2007 Lloyd-Jones 2009) New York State (NYS) and other northeast states have high IHD and low stroke mortality with NYS having the highest IHD mortality and lowest stroke mortality among the 50 states (Howard et al 2009)

CVD Data Sources CVD mortality rates are more easily measured than incidence prevalence or hospitalizations because mortality data from vital records are a comprehensive and relatively accessible data source Information for estimating the incidence and prevalence of CVD outcomes comes primarily from health interview physical examination and record surveys Hospitalization data are available for NYS and the US but are generally collected and analyzed in order to track trends in hospital care and resources expended Actual CVD incidence data validated by physician diagnosis physician examination or reviews of medical records are not directly available for the entire US or for specific geographic areas of NYS Such information is available from population-based cohort studies that follow people over time in selected communities While the concern underlying this research project is to understand CVD levels in the community ie incidence and prevalence we do not have data that directly provide us with this information In the absence of such data hospitalizations are frequently used as indicators of the burden of disease in communities particularly in surveillance or hypothesis-generating investigations such as this one In NYS the Behavioral Risk Factor Surveillance System (BRFSS) an ongoing monthly telephone survey of adults aged 18 years and older provides information about the burden of CVD by age sex race income and education

CVD Risk Factor Data The BRFSS data for NYS show that over 20 of the population aged 65 and older reported having been diagnosed with CVD versus less than five percent of the population under 65 Males were slightly more likely to report having CVD than females Nonshyhispanic whites compared to other ethnic groups combined were slightly more likely to report having CVD None of these slight differences were statistically significant however The only statistically significant demographic risk factors in the BRFSS data were low income and education levels Statistically significant elevations of reported CVD were observed for respondents with incomes below $25000 and respondents with a high school degree or less education (NYS DOH 2009a)

Another factor that can affect the outcome measure hospitalizations is the likelihood of mortality prior to reaching the hospital Studies show a variety of factors affect whether an individual experiencing a heart attack or stroke outside the hospital survives to be counted in hospitalization data These factors include lack of knowledge about personal risk or prior disease socio-economic status race marital status and distance to the hospital (Galea 2007

4

Barnett 2006 Ayala 2003) Population-based studies report sudden cardiac death usually outside the hospital to be the first overt manifestation of heart disease for 40 to 60 of all cases ( Kannel 1987 deVreede-Swagemakers 1997 UK Heart Attack Study 1998) An analysis of out-of-hospital mortality from AMI in Pennsylvania in 1998 showed that in relatively rural areas where travel time to the hospital was 25 minutes or longer 72 of all AMI deaths occurred outside the hospital compared with a statewide average of 49 of AMI deaths occurring outside of the hospital (OrsquoNeill 2003) In New York State 55 of IHD deaths occur outside the hospital compared to 35 of stroke deaths (NYS DOH 2009b)

Hudson River PCB Contamination

During an approximate 30-year period ending in 1977 the General Electric (GE) plants in Hudson Falls and Fort Edward used PCBs in the manufacture of electrical capacitors and discharged large quantities of PCBs into the upper Hudson River Estimates of the total direct discharges to the river are as high as 1330000 pounds In 1983 the US EPA classified a 200shymile section of the Hudson River from Hudson Falls to New York City as a National Priority List site making it one of the largest Superfund sites in the United States In 2002 the US EPA issued a Record of Decision calling for targeted environmental dredging and removal of PCB-contaminated sediment in the upper River (USEPA 2002) The dredging began in May 2009

The Superfund environmental investigations and planned remedial actions have focused on the areas of the river near Hudson Falls and Fort Edward where the highest levels of PCBs in sediments are located The 2002 Record of Decision provides information about the human health risk assessment conducted for the Upper and Mid-Hudson and the ecological risk assessment conducted for the Upper and Lower Hudson River (USEPA 2002) Fish contamination and fish consumption are the primary concerns at the site

In 1976 fishing was banned in the upper river and in 1995 a catch and release policy replaced the ban NYS DOH continues to recommend people eat no fish from the Upper Hudson River that children under age 15 and women of childbearing age eat no fish from the entire 200 mile stretch of river below Hudson Falls and that the general population eat none of most species of fish caught between the federal dam at Troy and Catskill approximately 40 miles south of Troy The Lower Hudson remains closed to commercial fishing for striped bass and eight other species

The fishing bans and advisories were issued because ingestion of fish is the major potential pathway for exposure for the general population Many previous studies have shown relationships between consumption of fish from contaminated waters and human PCB levels Another potential pathway for exposure less frequently studied to date is inhalation of PCBs NYS DOH and ATSDR launched ldquoThe Hudson River Communities Projectrdquo in 2000 to address this issue (More information on this project is provided below)

POPS and PCBs

The category ldquopersistent organic pollutantsrdquo (POPs) includes man-made chemicals that are by definition persistent in the environment Examples of POPs are polychlorinated biphenyls

5

(PCBs) polychlorinated dibenzo-p-dioxins (PCDDs) polychorinated dibenso-furans (PCDFs) and pprsquo-diphenyldichloroethene (DDE) a breakdown product of the pesticide dichlorodiphenyltrichloroethane (DDT) These chemicals are stored in human fat tissues accumulating over an individualrsquos lifetime They also accumulate in fish and other species thereby concentrating up the food chain Levels of most POPs measured in blood serum of Americans have decreased in the last few decades An exception is polybrominated diphenyl ethers (PBDEs) levels of which have been increasing in recent years While levels of many POPs are declining a large number continue to be detectable in blood for most of the US population (USDHHS 2009)

This follow-up investigation focuses on the Hudson River where the POPs of concern are PCBs PCBs are a group of synthetic chlorinated compounds manufactured in the US through 1977 Production was halted due to concerns about the persistence of PCBs in the environment and the toxicity of PCBs shown in animal models PCBs were used primarily by the electrical industry for capacitors and transformers and were also used in hydraulic fluids fluorescent light fixtures flame retardants inks adhesives carbonless copy paper paints pesticide extenders plasticizers and wire insulators PCBs can be oily liquids or solids and some are volatile and can exist as vapor in air PCBs in the environment are generally mixtures of the many different types known as congeners (ATSDR 2000)

The variety of types of PCBs adds to the complexity of studying adverse animal and human health effects Increasingly PCB research focuses on specific PCB congeners and groups of congeners because specific types of congeners appear to have different mechanisms of action and toxicity Dioxin-like PCBs (DLPCBs) are an important subset of PCBs due to concerns about their toxicity The focus on dioxin-like PCBs is paralleled by increasing research on the group of POPs with dioxin-like properties This type of PCB and POP research assigns toxic potency values relative to dioxin to various PCBs and POPs to provide an overall toxic equivalency value to specific PCB and other POPs

From 2002 through 2007 NYS DOH and SUNY Albany School of Public Health researchers conducted the Hudson River Communities Study among residents of Hudson Falls and Fort Edward the villages where the PCB discharges from two GE electrical capacitor plants occurred The comparison community was upriver from the discharges in Glens Falls This studyrsquos evaluation of exposures included measuring PCBs in outdoor air and in blood serum The study found that PCB levels in outdoor air were statistically significantly higher in the Hudson FallsFort Edward area (072 nanograms per cubic meter) than in the upriver Glens Falls area (040 nanograms per cubic meter) but that the PCB concentrations in the Hudson FallsFort Edward area were lower than those in other communities with known PCB-contaminated sites and similar to levels reported in other locations in the northeastern United States A statistically significant reduction in the total PCB concentrations in outdoor air samples was observed at a distance of 1200 meters (34 mile) from the river from 0760 ngm3 to 0497 ngm3 (Fitzgerald et al 2007 Palmer et al 2008)

The average PCB concentrations in serum for the study and comparison populations did not differ significantly (307 ppb wet weight versus 323 ppb) The average toxic equivalency quantities (TEQ) for dioxin-like PCB congeners for the study and comparison area populations

6

also did not differ In addition the study found no detectable differences in study subjectsrsquo serum PCB levels associated with proximity to contaminated portions of the river or other PCB sources The study did find that lifetime consumption of Hudson River fish was associated with higher serum PCB levels (Fitzgerald 2007 Palmer 2008) Another recent biomonitoring study of Brooklyn Manhattan and New Jersey fisherman who consumed versus those who did not consume their catch from the lower Hudson River and New YorkNew Jersey harbor suggests eating fish from these waters is not associated with an increased body burden of PCBs or most other organochlorines This study did not address the inhalation pathway (Morland 2008)

PCBs and Cardiovascular Disease Outcomes

Research studies on adverse health effects associated with PCB exposure have focused on many types of outcomes organs and biological systems including cancer liver reproductive and developmental effects thyroid endocrine immune system and neurological effects as well as specific effects on the skin and eyes Relatively little research has focused to date on associations between PCB exposures and CVD outcomes The majority of human studies have assessed health effects in occupational groups having relatively high-level exposures among people consuming fish from contaminated waters or people exposed to relatively high levels due to accidental contamination of food (USDHHS NIH 2010)

PCBs have been shown to produce a variety of toxic effects in animals These effects include cancer allergies hypersensitivity damage to the central and peripheral nervous systems reproductive disorders and disruption of the immune system (ATSDR 2000) In contrast research to date does not tell us whether specific types of human health effects have occurred from the low-level exposures experienced by the general population Some human epidemiological studies of relatively low levels of PCDDs PCDFs and PCBs suggest possible relationships not only with neurodevelopment of infants but also diabetes and a condition called insulin resistance conditions that increase CVD risk (Longnecker 2001 Carpenter 2006 Arisawa 2005)

In 1998 a study of a population in Seveso Italy exposed to relatively pure TCDD (dioxin) from an industrial accident showed increased mortality from CVD respiratory diseases and diabetes (Bertazzi 1998) These findings suggested the possibility that other dioxin-like chemicals such as dioxin-like PCBs might increase risk for these diseases Some more recent animal studies described below suggest possible mechanisms of action for PCB effects that increase risk for CVD diabetes and hypertension Another relevant source of information is from research using the biological monitoring data from the National Health and Nutrition Examination Survey (NHANES) project to look at the association between PCB levels in blood and self-reported diabetes and heart disease The body of evidence based on analytical human epidemiological studies remains relatively sparse however

Some recent studies using NHANES biomonitoring data show associations between PCB levels and CVD (Lee 2006 Lee 2007 Ha 2007 Everett 2008) These studies are cross-sectional (comparisons at one point in time) evaluations of PCB levels measured in blood samples from individuals and self-reported CVD and diabetes a risk factor for CVD While these studies use individual-level biomonitoring and self-reported health outcome data they are considered to be

7

relatively weak in terms of providing evidence of causality because both exposure and outcome are measured at the same point in time Given the nature of NHANES data collection and reporting practices these studies are not able to control for possible geographical confounding which may occur if urban versus rural residence is associated with higher PCB levels and higher risk for CVD for example The higher PCB levels among African Americans in these data suggest urban residence may be playing a role in the PCB data

The results of animal studies suggest multiple mechanisms by which PCB exposures could affect cardiovascular risk as well as multiple types of adverse health outcomes One hypothesized mechanisms of action specifically for dioxin-like PCBs is that they may have anti-estrogenic effects which impact CVD risk factors such as serum cholesterol levels and blood pressure (Lind 2004) Several studies suggest PCBs interact with a specific receptor (AhR) and increase cellular oxidative stress causing inflammation cell injury and cardiovascular dysfunction (Hennig 2002) Similarly some studies suggest PCB-induced inflammation of fat cells leads to obesity and increased risk for coronary artery disease (atherosclerosis) (Arsenescu 2008)

3 FOLLOW-UP STUDY DATA SOURCES AND METHODS

This follow-up investigation will provide perspective on the 2004-2005 findings (Sergeev and Carpenter 2005 Shcherbatykh et al 2005) by producing results from analyses conducted at a more detailed level of geography the block group rather than ZIP code in order to better control for confounding When the term confounding is used by epidemioogists it refers to the problem that can arise when there is a factor such as income level that is not the factor of interest in the study but that is related to the risk factor being studied (residential location ndash proxy for exposure) and to the disease being studied (CVD) This type of factor is labeled a ldquoconfounderrdquo because it can confound ie interfere with or confuse the evaluation of associations among the variables of interest in the data All types of epidemiological studies face issues of confounding but ecological ie group-level analyses face additional challenges related to potential confounding because analyses can not account for some important individual differences within the groups that are the units of analysis (Morganstern 1998b Hertz-Picciotto 1998) This investigation was able to control for each individualrsquos age sex and race but not for individual-level risk factors such as income or education level

The follow-up investigationrsquos specific objectives are to

improve the ability to control for confounding by socio-economic status through the use of block groups rather than ZIP codes as the unit of analysis

compare the hospitalization rates among residents of block groups nearest the Hudson River to those among residents of block groups farther from the Hudson River

compare the findings of this follow-up study to those of the earlier studies

Study Area and Exposure Areas

The overall study area is comprised of the 12 counties with geographic areas abutting the Hudson River south of the source of PCB contamination from the General Electric facility in Hudson Falls The 12 counties are Westchester Rockland Putnam Orange Dutchess Ulster

8

Columbia Greene Rensselaer Albany Washington and Saratoga US Census boundaries are used to define the study area because census population estimates provide the data for the study area populationrsquos size and the populationrsquos age sex and race distribution Because the study period spanned 1990 through 2005 and populations change over time 1990 and 2000 Census data were used to linearly interpolate the between-year populations from 1991 through 1999 for each block group 2000 data were used without extrapolation to estimate block group populations from 2001 through 2005

Because block group boundaries change between the Census years 1990 block populations were normalized to fit 2000 block group boundaries before the estimation of sex age and race group specific populations In the digital map 2000 block group boundaries were overlaid on the 1990 block centroids to assign 1990 block population counts to the corresponding 2000 block group boundaries Then 1990 block populations were summed to fit the block group boundaries Next population counts by age group (0-14 15-24 25-34 35-44 45-54 55-64 65-74 75+) sex (male female) and race (white black asian other) were estimated for each year at the block group level This investigation includes all ages over 25 years and the study population is divided into four race categories white black or African-American Asian and ldquoOtherrdquo The Pacific Island group was combined with the Asian group and American Indian ldquootherrdquo and ldquomultiple racesrdquo categories are combined into a larger ldquootherrdquo category

Classification of the population into potential exposure categories for this study is based exclusively on location relative to the Hudson River Two types of exposure proxy categories were developed block group adjacency and block group centroid distance to the river shore For the adjacency categories block groups were assigned to the adjacent category if they directly abutted the Hudson River and all other block groups in the 12 counties were assigned to the nonshyadjacent category

For the distance categories a central point (centroid) within each block group was identified and used to assign block group centroid distance to the river shore The block group centroids were population-weighted which means the geographic distribution of households within 2000 census blocks were used to locate block group centroids based on the geographic population distribution By locating a population-weighted block centroid rather than a more straightforward distance centroid the assignment of block group distance to the river shore more accurately reflects distance to the river for the majority of block group residents Buffers of one-half mile and one mile from the Hudson River boundary were created for the assignment of each block group into one of the three distance categories less than frac12 mile (closest) frac12 to one mile (close) and farther than one mile from the river (distant)

Distance categories were selected a priori at the start of the study Because environmental sampling data were not collected along the length of the river included in this investigation we are not able to determine whether these distance categories represent actual differences in exposures As discussed previously a statistically significant reduction in the total PCB concentrations in outdoor air samples was observed at a distance of 1200 meters (34 mile) from the river from 0760 ngm3 to 0497 ngm3 in a study conducted in the Hudson Falls Fort Edward area near the source of PCBs (Fitzgerald 2007 Palmer 2007)

9

For the adjacency categories of the 2371 block groups in the study area 255 (11) were classified as adjacent to the Hudson River and 2116 (89) were classified as non-adjacent (Figure 2) For the distance categories 225 (9) were classified as closest to the river 216 (9) as close and 1930 block groups (81) as distant from the river

Comparison of Figures 1 and 2 show that block groups include smaller geographic areas than ZIP codes and that block groups are relatively smaller in area in urban versus rural areas The study area includes a total of 368 ZIP codes compared to a total of 2371 block groups ZIP code average population in the study area is approximately 8000 compared to block group average population of 1200 Comparison of Figures 2 and 3 shows that using population-weighted centroid distance to the river produces different exposure areas than adjacency Fifty-three (20) of the adjacent block groups were categorized for the distance indicator as distant because the population in the block group was concentrated in the area farther than one mile from the river On the other hand about 11 of the 2116 non-adjacent block groups were classified as closest (77) or close (162)

Hospitalization Data

Hospitalizations for ischemic heart disease (IHD) and cerebrovascular disease (stroke) were included in this study More specifically hospital discharge records that listed as the principal diagnosis either IHD (ICD-9-CM 410-414 and 4297) or cerebrovascular disease (ICD-9-CM 430-436) and subsets of each of these categories similar to the subsets evaluated in the Sergeev et al and Shcherbatykh studies were included The subsets are AMI and angina subsets of IHD and ischemic and hemorrhagic stroke which are subsets of cerebrovascular disease (stroke) (Principal diagnosis represents the illness for which the person was admitted to the hospital) Angina is an outcome not evaluated in the prior studies It was added here for comparative purposes because it represents an outcome appropriately addressed by primary outpatient rather than inpatient hospital care Angina is one of a group of outcomes included in the recently developed national ldquoPrevention Quality Indicatorsrdquo (USDHHS AHRQ 2007) as an indicator for geographic areas or populations for whom primary care is inadequate Table 3 lists the specific ICD-9-CM codes and definitions included in this investigation

While one of the previously published studies (Huang et al 2006) showed elevations in hypertension (high blood pressure) rates this outcome is rarely listed as the principal reason for hospitalization As a secondary diagnosis however hypertension is extremely common with substantial overlap with the primary diagnoses of IHD and stroke In addition hypertension hospitalizations were not as elevated in the ZIP codes adjacent to the Hudson River as the IHD and stroke outcomes As a practical matter geocoding the hypertension hospitalization data (1364000 hospitalizations) would overwhelm the resources for this project For these reasons hypertension was not included in this follow-up investigation

The source of the hospitalization data was the NYS DOH Statewide Planning and Research Cooperative System (SPARCS) established in 1979 to collect detailed records on discharges from hospitals located in New York State Persons seen in the emergency department but not admitted are not included in the SPARCS data Hospitalizations at US Veterans Administration

10

Hospitals are also not included in this dataset We identified the hospital transfers in the data and excluded them so that individuals are only counted once for the same event

The NYS DOH Data Protection Review Board for SPARCS hospitalization data approved our use of hospitalization records for this investigation and provided approximately 460000 records with a principal diagnosis of either IHD or stroke Because of the large number of cardiovascular hospitalizations and this studyrsquos requirement for geocoding each hospitalized personrsquos address we selected a 50 random sample of the hospital discharges with principal diagnoses of IHD or stroke for inclusion in the study (229948 records) Approximately one percent of these records (2737) were eliminated from the study because they contained no residential address resulting in 227211 hospitalization records 24939 of these records were eliminated after geocoding because they were not within the 12 county study area This left 202272 records remaining in the study

For geocoding to the block group level the hospitalization street addresses were modified as needed using US Postal Service standards The addresses were assigned geographic coordinates using commercially available geocoding software (MapMarker 2004) A combination of land parcel data and street centerline files were used to assign geographic coordinates No contact was made with cases parents legal guardians or next of kin of cases to determine residential locations

Once geographic coordinates were assigned to cases through address-matching the case locations were overlaid onto digital maps of the study area using a geographic information system so that the number of observed cases falling within the study area boundaries could be determined and a 2000 Census block group could be assigned to each hospitalization In order to protect confidentiality no maps of individual case locations are provided Geographic coordinates were not able to be assigned for some addresses for example rural routes or post office boxes Internet searches were conducted for addresses that were located in facilities such as nursing homes and trailer parks

While fewer than one percent of records were missing sex or age information approximately 15000 records seven percent were missing race information Evaluation of the geographic and temporal distribution of these records showed they were not randomly geographically distributed 47 of the records with no race information were from one hospital for that hospital race was missing from 100 of the records in all years except one after 1997 mid-way through the study time-frame Evaluation of other records missing race showed similar patterns of 100 of records missing race in specific years In order to prevent biased hospitalization rate estimates records missing race were assigned race randomly assuming a race distribution equal to that of the individualrsquos residential block group

Statistical Analysis

This investigation used negative binomial multivariable regression models the statistical methods used in the previous studies (Sergeev and Carpenter 2005 [IHD] Shcherbatykh et al 2005 [stroke]) The follow-up studyrsquos analyses compared hospitalization rates for populations in block groups adjacent or in close proximity to the Hudson River to rates for populations more

11

distant from the Hudson River while adjusting for the potential influence of other factors Sex age and race-specific hospitalization count data comprise the dependent variable and the statistical methods attempt to predict the counts using the sex age and race distributions from Census data as well as block group-level demographics such as median income Models were first estimated using basic Census information for age sex and race Then more fully adjusted models were estimated using information at the block group level for income education population density Hispanic ethnicity and distance to the nearest hospital

More specifically each hospitalized personrsquos block group was used to assign quartile categories of block group median household income of population with less than a college education population density of the population identified as Hispanic and distance from the block group centroid to the nearest hospital These variables were selected in advance The population density variable was selected to account for potential effects on hospitalization rates of large differences in population density from the relatively rural northern counties to the more urban southern counties in the study area of the population identified as Hispanic was included because Hispanic ethnicity may be associated with CVD risk but Hispanic ethnicity was not available in the hospitalization records Distance to the hospital was included because studies show that shorter distances increase the likelihood that individuals with AMI and stroke survive to be admitted to the hospital

As described previously the literature about risk factors for CVD shows strong associations between lower income and education and higher CVD risk Regarding the effects of income and education on hospitalization rates the literature from analytic studies specifically of hospitalizations for CVD is quite sparse Most analytic studies focus on prevalence incidence or mortality in population-based studies or cohort studies and there are not many of these An example of a study that was useful as background is a study of income level and CVD mortality that showed strong associations of lower income and higher CVD mortality for men and women with the relationship for men being linear throughout the range of income levels while for women the relationship was only strong at levels of income below the median (Rehkopf 2008)

This follow-up studyrsquos approach was to use the same set of variables for the more fully adjusted models using quartile cut points for all the sexrace groups and types of CVD diagnostic subsets The 2000 Census data were used to create the population-weighted quartile categories to be used in the statistical analyses The quartile cut points are shown below

population density(persons per square mile)

median 1999 household income

with less than a college education

with Hispanic ethnicity

Hospital distance

Quartile 1 61 - lt507 $2499-lt$40968 0 - lt31 0 - lt2 lt25 (median) Quartile 2 507 - lt102 $40968 - lt54650 31 - lt43 2 - lt5 25 - lt6 (50-75) Quartile 3 102 - lt6410 $54650 - lt72944 43 - lt55 5 - lt10 6 ndash 11 (75-90) Quartile 4 6410 - 94372 $72944 - 200001 55 - 100 10 - 86 11 ndash 45 (90-100)

SAS version 9 provided the statistical programs used to estimate hospitalization rates for exposure proxy categories and demographic categories The models provide ldquocontrast

12

estimatesrdquo ie rate ratios for one category versus a reference category of a variable and 95 confidence intervals (CI) for these rate ratios For the location categories rate ratios for the areas adjacent are compared to areas not adjacent to the river (reference) and areas closest and close to the river are compared to areas distant from the river (reference) If a rate ratio is greater than 100 the model has estimated an elevation of hospitalizations in that category of the study population compared to the reference population If the rate ratio is less than 100 then there are fewer than expected hospitalizations compared to the reference population

The magnitude of the elevation or deficit is also estimated by the rate ratio For instance if the hospitalization rate is twice as high in the adjacent population it would result in a rate ratio of 200 while a 50 higher rate would result in a rate ratio of 150 If the hospitalization rate in the comparison group was only half as big as in the reference group this would result in a rate ratio of 050 The 95 confidence interval assists with determining whether the observed differences are statistically meaningful ie statistically significant The confidence interval provides the range in which there is a 95 probability of including the true rate ratio

In addition to directly assess potential confounding we examined the associations between income and exposure proxy categories and between income and CVD hospitalization risk Because the modeling process provides rate ratios that provide comparisons among groups but not the rates themselves we also calculated the rates per population for each sex race and income category in order to directly observe the relative magnitudes of CVD hospitalization rates Finally analyses stratified by income quartile were also conducted in order to assess the potential role of confounding by income The stratified analyses consist of negative binomial regressions conducted separately for each quartile of income

4 RESULTS

Demographics Associations with Proximity to the River

Study Population Table 4 provides detailed demographic characteristics for the total study population and the populations of the adjacent versus non-adjacent and nearer versus farther from the river block groups from the 2000 Census The total population of the study area for this one year of the study time period is 2924631 persons The first row of Table 4 shows that approximately 11 percent of the study area population is classified as living in block groups adjacent to the Hudson River and 89 live in non-adjacent block groups Approximately 8 of the studyrsquos population lives in the closest block groups approximately 10 lives in close block groups and approximately 82 in distant block groups The table also shows that approximately 81 of the total study population is identified as white 9 black 3 Asian less than 01 Pacific Islander 02 American Indian 4 ldquootherrdquo and 2 ldquomultiple racesrdquo

The 2000 Census data (Table 4) also show that the populations in block groups adjacent and nonshyadjacent to the river are similar in terms of the percent of minority population but the adjacent block groups include a smaller percentage of persons who have attended college and a higher percentage of households below the poverty level Median income is $47320 for block groups adjacent versus $62668 for block groups not adjacent to the river Data for the study area from the 1990 Census show similar distributions (data not shown)

13

For the distance to the river categories there is a gradient for the three levels of distance from the river with the closest block groups on average having higher percentages of minorities smaller percentages of persons who have attended college and higher percentages of households below the poverty level (Table 4) The populations of the close and closest block groups are more similar to each other than to the populations of the distant block groups The population of the closest block groups is estimated to be 19 black close block groups 14 black and distant block groups 8 black The proportions for the ldquootherrdquo category used in the analyses which includes multiple races and American Indians are approximately 12 in both the closest block groups and in the close block groups compared to 5 in the block groups most distant from the river Median income for the closest block groups is $40748 close block groups $44962 and most distant block groups $64847

The bar chart below illustrates the correlations between US Census 2000 indicators of block group socio-economic status and block group centroid (population weighted) distance from the river for the study population

2000 Census Demographics for Study Population by Distance to River Category

black other hispanic no college below poverty

of pop ulati on

0

10

20

30

40

50

60

lt12 mile

12 to 1 mile

gt1 mile

Table 5 shows selected demographic information for the person-year estimates used in the study analyses for the 1990-2005 study period Person-year data represent individuals from Census estimates counted repeatedly for each year of the study timeframe As described in the Methods section the estimation of person years used interpolation to account for population changes from 1990-2000 and techniques for taking account of block boundary changes The person-years estimates for demographic categories (complete data not shown) show similar patterns as the 1990 Census data (data not shown) and 2000 Census data (Table 5) Table 5 for example shows nearly identical percentages of population in each of the river distance categories as Table 4

Table 5 also shows person year information for income quartiles and exposure proxy categories as well as race and exposure proxy categories that is consistent with the Table 4 data for the

14

2000 population Table 5 row percentages show for example that while approximately 11 of the total study population live in adjacent block groups approximately 19 of the population in the lowest income quartile live in adjacent block groups By design approximately 25 of the total study population is in each of the income quartiles Table 5 column percentages show however that in the adjacent block groups 41 of the population is in the lowest income quartile and only 9 are in the highest income quartile

For the block groups within frac12 mile of the river the distribution among income quartiles is even more skewed towards the lowest income quartile 61 of the population living within frac12 mile of the river is in the lowest income quartile and only 6 is in the highest income quartile Given that approximately 25 of the total study population is within each income quartile these percentages indicate that living within the block groups closest to the river is associated with more than a doubling of the likelihood of being in a block group in the lowest socio-economic status quartile 436 of those living in the block groups close to the river (12 to one mile from the river) are in block groups in the lowest income quartile While less than a doubling this indicates that living in the close block groups is associated with a 75 increase in the likelihood of being in the lowest quartile for socio-economic status

Hospitalizations Table 6 shows the process of removing hospitalization records from the study for a variety of reasons The first column includes the 202272 records that contained a residential address and were located within the 12-county study area as described in the Methods section The next five columns show the records remaining after data cleaning exclusion of transfers elimination of records with missing covariates and exclusions due to poor geocoding results Approximately 33000 records were eliminated because they were transfers from one hospital to another for the same event For approximately 1100 records geocoding was inadequate for assigning block group for imputing race as described in the Methods section 336 records were missing sex (11) or age (325) Of the 167733 records remaining 154220 (92) were able to be adequately geocoded to a street address for assignment in a specific block group While approximately eight percent of hospitalizations were eliminated due to inadequate geocoding Table 6 shows that the percentages of hospitalization records for white black Asian and ldquootherrdquo categories and the distribution among diagnostic categories change very little after these records were dropped

The final column of Table 6 shows that of the 154220 hospitalizations in the analysis 101012 IHD hospitalizations comprise the largest health outcome category and are 65 of the hospitalizations in the study AMI hospitalizations (35811) account for 35 of the IHD hospitalizations and angina (4731) accounts for five percent of the IHD hospitalizations Stroke hospitalizations (53208) comprise 34 of the studyrsquos hospitalizations Approximately 36 of the stroke hospitalizations are for ischemic stroke (19371) and approximately 11 are for hemorrhagic stroke (5995)

Table 7 shows the distributions of hospitalizations among the four income quartiles While each income quartile (by definition) contains approximately 25 of the study population for all types of hospitalizations we see that the lowest income quartile includes more than 25 of hospitalizations from a minimum of 286 (IHD) to a maximum of 356 of hospitalizations (angina) The highest income quartile includes fewer than 25 of hospitalizations for each CVD

15

category ranging from 182 for angina to 224 for hemorrhagic stroke These deviations from the 25 quartile distribution show a strong linear pattern with the largest percentage of hospitalizations coming from residents of block groups in the lowest income quartile and the smallest percentages coming from residents of the most wealthy block groups For AMI the lowest income quartilersquos share of hospitalizations is 50 higher than the highest income quartilersquos share of hospitalizations (30 vs 20) IHD and stroke show similar differences and angina shows a doubling with the lowest income quartile contributing 36 of hospitalizations compared to the highest income quartilersquos 18 share of hospitalizations

Table 7 also shows the distribution of hospitalizations among the race categories and the study population race category distribution for comparison Blacks show lower percentages of IHD and AMI than their population percentages and higher percentages of angina and stroke While full presentation of race and sex findings is beyond the scope of this report the lower than expected percentages of IHD and AMI among blacks who on average have lower socioshyeconomic status and are at greater risk for CVD hospitalizations raised questions that were addressed by calculating hospitalization rates per 1000 population for each sex and race group for each income quartile (Appendix B Table 2) Figures 4 5 and 6 show these rates for the white black and Asian categories (These rates for the ldquootherrdquo category are comparatively very high and are not shown on these graphs This issue is addressed in the Discussion section) Figures 4 5 and 6 show the consistent and strong effects of income quartile on hospitalization rates for AMI and stroke for whites blacks and Asians of both sexes The other outcomes show similar patterns (The ldquootherrdquo group showed an opposite pattern [See Discussion Section])

Demographics Summary

The strong associations of lower income with residence closer to the river (Table 5) and lower income with higher hospitalization rates (Tables 7 amp 8) create a difficult challenge for this investigation The multivariable analyses of hospitalization rates attempt to control for socioshyeconomic differences However because these income differences which strongly affect hospitalization rates are also strongly correlated with the gradations of exposure (river distance) it will be difficult to disentangle the influence of socio-economic status from the influence of potential exposures

Statistical Analyses Proximity to the River and CVD Hospitalization Risk

All the variables described above were included in the multivariable regression models and almost every quartile category for every variable showed statistically significant and expected types of effects for all the CVD hospitalization categories (Appendix B Tables 1a and 1b) The following presentation of results focuses on the rate ratios for the exposure proxy variables Table 8a (IHD) and 8b (stroke) show the regression modeling results Results are shown first for models that include only age sex and race in addition to the exposure proxy categories and second for models that also include median income education population density Hispanic ethnicity and distance to nearest hospital The tables provide numbers of hospitalizations and person-years to assist with interpretation For example for total IHD we can see that while 108 of the study population lives in adjacent block groups 122 of the hospitalizations are among people in adjacent block groups This difference of 14 percentage points is the basis for

16

the finding of a statistically significantly elevated rate ratio (risk) for IHD among people living in the adjacent block groups (RR=133 CI 112-157)

Looking across the columns of Tables 8a and 8b we can see the rate ratios for each of the outcomes for the adjacent versus non-adjacent block groups and the results are similar to those for IHD The elevated rate ratios range from 123 for angina to 136 for AMI The only rate ratio not elevated in these first models adjusting only for age sex and race is for hemorrhagic stroke Going down the rows of the Table the next set of rate ratios are for the distance to the river categories The findings are very similar to those for the adjacent block groups Here however we have two exposure proxy categories allowing comparison of the two These rate ratios show no evidence that living in the block groups closest to the river (lt12 mile to the river) produces greater risk than living in the block groups close to the river (12 to 1 mile from the river)

The next set of rate ratios are for models with additional adjustment for the demographics of each hospitalized personrsquos block group as well as distance to the nearest hospital As expected adjustment for block group median income and other demographics accounts for some of the variability in hospitalization counts and thereby reduces the magnitude of the rate ratios estimated for the river adjacency and distance categories The rate ratio elevation for total IHD for example is reduced from 133 to 107 (RR=107 CI 104-111) The elevated rate ratios in these more fully adjusted models for the adjacency analyses are of similarly small magnitude ranging from 106 for AMI to 110 for ischemic stroke Four of the six outcomes show statistically significant elevations Again only hemorrhagic stroke shows no evidence of an elevated rate ratio

For the river distance categories none of the rate ratios show any evidence that living in the block groups closest to the river (lt12 mile to the river) produces greater risk than living in the block groups close to the river (12 to 1 mile) Of the six types of outcomes only two show statistically significantly elevated rate ratios for living in the closest block groups (total stroke and ischemic stroke) For the close block groups four out of six outcomes show statistically significantly elevated rate ratios The statistically significantly elevated rate ratios vary from 106 (AMI stroke) to 113 (angina)

To address the strong evidence described earlier for confounding in these data ie the association of lower income with both CVD hospitalizations and residence closer to the river we also conducted the analyses separately for each quartile of income These results are presented in Table 9 The first results are from analyses adjusting only for age sex and race and the second set of results are from models including the additional demographic factors For simplicity rate ratios are shown only for block groups closest to the river compared as in the other models to block groups farther than one mile from the river (Results were similar for the other distance categories (data not shown)

For the two lowest income quartiles Table 9 shows results similar to those for the study population as a whole (Table 8a 8b) However for the two highest income quartiles which represent frac12 of the study population Table 9 shows a consistent pattern of reduced rate ratios for residence in the closest block groups within frac12 mile of the river In the highest income quartile

17

Barnett 2006 Ayala 2003) Population-based studies report sudden cardiac death usually outside the hospital to be the first overt manifestation of heart disease for 40 to 60 of all cases ( Kannel 1987 deVreede-Swagemakers 1997 UK Heart Attack Study 1998) An analysis of out-of-hospital mortality from AMI in Pennsylvania in 1998 showed that in relatively rural areas where travel time to the hospital was 25 minutes or longer 72 of all AMI deaths occurred outside the hospital compared with a statewide average of 49 of AMI deaths occurring outside of the hospital (OrsquoNeill 2003) In New York State 55 of IHD deaths occur outside the hospital compared to 35 of stroke deaths (NYS DOH 2009b)

Hudson River PCB Contamination

During an approximate 30-year period ending in 1977 the General Electric (GE) plants in Hudson Falls and Fort Edward used PCBs in the manufacture of electrical capacitors and discharged large quantities of PCBs into the upper Hudson River Estimates of the total direct discharges to the river are as high as 1330000 pounds In 1983 the US EPA classified a 200shymile section of the Hudson River from Hudson Falls to New York City as a National Priority List site making it one of the largest Superfund sites in the United States In 2002 the US EPA issued a Record of Decision calling for targeted environmental dredging and removal of PCB-contaminated sediment in the upper River (USEPA 2002) The dredging began in May 2009

The Superfund environmental investigations and planned remedial actions have focused on the areas of the river near Hudson Falls and Fort Edward where the highest levels of PCBs in sediments are located The 2002 Record of Decision provides information about the human health risk assessment conducted for the Upper and Mid-Hudson and the ecological risk assessment conducted for the Upper and Lower Hudson River (USEPA 2002) Fish contamination and fish consumption are the primary concerns at the site

In 1976 fishing was banned in the upper river and in 1995 a catch and release policy replaced the ban NYS DOH continues to recommend people eat no fish from the Upper Hudson River that children under age 15 and women of childbearing age eat no fish from the entire 200 mile stretch of river below Hudson Falls and that the general population eat none of most species of fish caught between the federal dam at Troy and Catskill approximately 40 miles south of Troy The Lower Hudson remains closed to commercial fishing for striped bass and eight other species

The fishing bans and advisories were issued because ingestion of fish is the major potential pathway for exposure for the general population Many previous studies have shown relationships between consumption of fish from contaminated waters and human PCB levels Another potential pathway for exposure less frequently studied to date is inhalation of PCBs NYS DOH and ATSDR launched ldquoThe Hudson River Communities Projectrdquo in 2000 to address this issue (More information on this project is provided below)

POPS and PCBs

The category ldquopersistent organic pollutantsrdquo (POPs) includes man-made chemicals that are by definition persistent in the environment Examples of POPs are polychlorinated biphenyls

5

(PCBs) polychlorinated dibenzo-p-dioxins (PCDDs) polychorinated dibenso-furans (PCDFs) and pprsquo-diphenyldichloroethene (DDE) a breakdown product of the pesticide dichlorodiphenyltrichloroethane (DDT) These chemicals are stored in human fat tissues accumulating over an individualrsquos lifetime They also accumulate in fish and other species thereby concentrating up the food chain Levels of most POPs measured in blood serum of Americans have decreased in the last few decades An exception is polybrominated diphenyl ethers (PBDEs) levels of which have been increasing in recent years While levels of many POPs are declining a large number continue to be detectable in blood for most of the US population (USDHHS 2009)

This follow-up investigation focuses on the Hudson River where the POPs of concern are PCBs PCBs are a group of synthetic chlorinated compounds manufactured in the US through 1977 Production was halted due to concerns about the persistence of PCBs in the environment and the toxicity of PCBs shown in animal models PCBs were used primarily by the electrical industry for capacitors and transformers and were also used in hydraulic fluids fluorescent light fixtures flame retardants inks adhesives carbonless copy paper paints pesticide extenders plasticizers and wire insulators PCBs can be oily liquids or solids and some are volatile and can exist as vapor in air PCBs in the environment are generally mixtures of the many different types known as congeners (ATSDR 2000)

The variety of types of PCBs adds to the complexity of studying adverse animal and human health effects Increasingly PCB research focuses on specific PCB congeners and groups of congeners because specific types of congeners appear to have different mechanisms of action and toxicity Dioxin-like PCBs (DLPCBs) are an important subset of PCBs due to concerns about their toxicity The focus on dioxin-like PCBs is paralleled by increasing research on the group of POPs with dioxin-like properties This type of PCB and POP research assigns toxic potency values relative to dioxin to various PCBs and POPs to provide an overall toxic equivalency value to specific PCB and other POPs

From 2002 through 2007 NYS DOH and SUNY Albany School of Public Health researchers conducted the Hudson River Communities Study among residents of Hudson Falls and Fort Edward the villages where the PCB discharges from two GE electrical capacitor plants occurred The comparison community was upriver from the discharges in Glens Falls This studyrsquos evaluation of exposures included measuring PCBs in outdoor air and in blood serum The study found that PCB levels in outdoor air were statistically significantly higher in the Hudson FallsFort Edward area (072 nanograms per cubic meter) than in the upriver Glens Falls area (040 nanograms per cubic meter) but that the PCB concentrations in the Hudson FallsFort Edward area were lower than those in other communities with known PCB-contaminated sites and similar to levels reported in other locations in the northeastern United States A statistically significant reduction in the total PCB concentrations in outdoor air samples was observed at a distance of 1200 meters (34 mile) from the river from 0760 ngm3 to 0497 ngm3 (Fitzgerald et al 2007 Palmer et al 2008)

The average PCB concentrations in serum for the study and comparison populations did not differ significantly (307 ppb wet weight versus 323 ppb) The average toxic equivalency quantities (TEQ) for dioxin-like PCB congeners for the study and comparison area populations

6

also did not differ In addition the study found no detectable differences in study subjectsrsquo serum PCB levels associated with proximity to contaminated portions of the river or other PCB sources The study did find that lifetime consumption of Hudson River fish was associated with higher serum PCB levels (Fitzgerald 2007 Palmer 2008) Another recent biomonitoring study of Brooklyn Manhattan and New Jersey fisherman who consumed versus those who did not consume their catch from the lower Hudson River and New YorkNew Jersey harbor suggests eating fish from these waters is not associated with an increased body burden of PCBs or most other organochlorines This study did not address the inhalation pathway (Morland 2008)

PCBs and Cardiovascular Disease Outcomes

Research studies on adverse health effects associated with PCB exposure have focused on many types of outcomes organs and biological systems including cancer liver reproductive and developmental effects thyroid endocrine immune system and neurological effects as well as specific effects on the skin and eyes Relatively little research has focused to date on associations between PCB exposures and CVD outcomes The majority of human studies have assessed health effects in occupational groups having relatively high-level exposures among people consuming fish from contaminated waters or people exposed to relatively high levels due to accidental contamination of food (USDHHS NIH 2010)

PCBs have been shown to produce a variety of toxic effects in animals These effects include cancer allergies hypersensitivity damage to the central and peripheral nervous systems reproductive disorders and disruption of the immune system (ATSDR 2000) In contrast research to date does not tell us whether specific types of human health effects have occurred from the low-level exposures experienced by the general population Some human epidemiological studies of relatively low levels of PCDDs PCDFs and PCBs suggest possible relationships not only with neurodevelopment of infants but also diabetes and a condition called insulin resistance conditions that increase CVD risk (Longnecker 2001 Carpenter 2006 Arisawa 2005)

In 1998 a study of a population in Seveso Italy exposed to relatively pure TCDD (dioxin) from an industrial accident showed increased mortality from CVD respiratory diseases and diabetes (Bertazzi 1998) These findings suggested the possibility that other dioxin-like chemicals such as dioxin-like PCBs might increase risk for these diseases Some more recent animal studies described below suggest possible mechanisms of action for PCB effects that increase risk for CVD diabetes and hypertension Another relevant source of information is from research using the biological monitoring data from the National Health and Nutrition Examination Survey (NHANES) project to look at the association between PCB levels in blood and self-reported diabetes and heart disease The body of evidence based on analytical human epidemiological studies remains relatively sparse however

Some recent studies using NHANES biomonitoring data show associations between PCB levels and CVD (Lee 2006 Lee 2007 Ha 2007 Everett 2008) These studies are cross-sectional (comparisons at one point in time) evaluations of PCB levels measured in blood samples from individuals and self-reported CVD and diabetes a risk factor for CVD While these studies use individual-level biomonitoring and self-reported health outcome data they are considered to be

7

relatively weak in terms of providing evidence of causality because both exposure and outcome are measured at the same point in time Given the nature of NHANES data collection and reporting practices these studies are not able to control for possible geographical confounding which may occur if urban versus rural residence is associated with higher PCB levels and higher risk for CVD for example The higher PCB levels among African Americans in these data suggest urban residence may be playing a role in the PCB data

The results of animal studies suggest multiple mechanisms by which PCB exposures could affect cardiovascular risk as well as multiple types of adverse health outcomes One hypothesized mechanisms of action specifically for dioxin-like PCBs is that they may have anti-estrogenic effects which impact CVD risk factors such as serum cholesterol levels and blood pressure (Lind 2004) Several studies suggest PCBs interact with a specific receptor (AhR) and increase cellular oxidative stress causing inflammation cell injury and cardiovascular dysfunction (Hennig 2002) Similarly some studies suggest PCB-induced inflammation of fat cells leads to obesity and increased risk for coronary artery disease (atherosclerosis) (Arsenescu 2008)

3 FOLLOW-UP STUDY DATA SOURCES AND METHODS

This follow-up investigation will provide perspective on the 2004-2005 findings (Sergeev and Carpenter 2005 Shcherbatykh et al 2005) by producing results from analyses conducted at a more detailed level of geography the block group rather than ZIP code in order to better control for confounding When the term confounding is used by epidemioogists it refers to the problem that can arise when there is a factor such as income level that is not the factor of interest in the study but that is related to the risk factor being studied (residential location ndash proxy for exposure) and to the disease being studied (CVD) This type of factor is labeled a ldquoconfounderrdquo because it can confound ie interfere with or confuse the evaluation of associations among the variables of interest in the data All types of epidemiological studies face issues of confounding but ecological ie group-level analyses face additional challenges related to potential confounding because analyses can not account for some important individual differences within the groups that are the units of analysis (Morganstern 1998b Hertz-Picciotto 1998) This investigation was able to control for each individualrsquos age sex and race but not for individual-level risk factors such as income or education level

The follow-up investigationrsquos specific objectives are to

improve the ability to control for confounding by socio-economic status through the use of block groups rather than ZIP codes as the unit of analysis

compare the hospitalization rates among residents of block groups nearest the Hudson River to those among residents of block groups farther from the Hudson River

compare the findings of this follow-up study to those of the earlier studies

Study Area and Exposure Areas

The overall study area is comprised of the 12 counties with geographic areas abutting the Hudson River south of the source of PCB contamination from the General Electric facility in Hudson Falls The 12 counties are Westchester Rockland Putnam Orange Dutchess Ulster

8

Columbia Greene Rensselaer Albany Washington and Saratoga US Census boundaries are used to define the study area because census population estimates provide the data for the study area populationrsquos size and the populationrsquos age sex and race distribution Because the study period spanned 1990 through 2005 and populations change over time 1990 and 2000 Census data were used to linearly interpolate the between-year populations from 1991 through 1999 for each block group 2000 data were used without extrapolation to estimate block group populations from 2001 through 2005

Because block group boundaries change between the Census years 1990 block populations were normalized to fit 2000 block group boundaries before the estimation of sex age and race group specific populations In the digital map 2000 block group boundaries were overlaid on the 1990 block centroids to assign 1990 block population counts to the corresponding 2000 block group boundaries Then 1990 block populations were summed to fit the block group boundaries Next population counts by age group (0-14 15-24 25-34 35-44 45-54 55-64 65-74 75+) sex (male female) and race (white black asian other) were estimated for each year at the block group level This investigation includes all ages over 25 years and the study population is divided into four race categories white black or African-American Asian and ldquoOtherrdquo The Pacific Island group was combined with the Asian group and American Indian ldquootherrdquo and ldquomultiple racesrdquo categories are combined into a larger ldquootherrdquo category

Classification of the population into potential exposure categories for this study is based exclusively on location relative to the Hudson River Two types of exposure proxy categories were developed block group adjacency and block group centroid distance to the river shore For the adjacency categories block groups were assigned to the adjacent category if they directly abutted the Hudson River and all other block groups in the 12 counties were assigned to the nonshyadjacent category

For the distance categories a central point (centroid) within each block group was identified and used to assign block group centroid distance to the river shore The block group centroids were population-weighted which means the geographic distribution of households within 2000 census blocks were used to locate block group centroids based on the geographic population distribution By locating a population-weighted block centroid rather than a more straightforward distance centroid the assignment of block group distance to the river shore more accurately reflects distance to the river for the majority of block group residents Buffers of one-half mile and one mile from the Hudson River boundary were created for the assignment of each block group into one of the three distance categories less than frac12 mile (closest) frac12 to one mile (close) and farther than one mile from the river (distant)

Distance categories were selected a priori at the start of the study Because environmental sampling data were not collected along the length of the river included in this investigation we are not able to determine whether these distance categories represent actual differences in exposures As discussed previously a statistically significant reduction in the total PCB concentrations in outdoor air samples was observed at a distance of 1200 meters (34 mile) from the river from 0760 ngm3 to 0497 ngm3 in a study conducted in the Hudson Falls Fort Edward area near the source of PCBs (Fitzgerald 2007 Palmer 2007)

9

For the adjacency categories of the 2371 block groups in the study area 255 (11) were classified as adjacent to the Hudson River and 2116 (89) were classified as non-adjacent (Figure 2) For the distance categories 225 (9) were classified as closest to the river 216 (9) as close and 1930 block groups (81) as distant from the river

Comparison of Figures 1 and 2 show that block groups include smaller geographic areas than ZIP codes and that block groups are relatively smaller in area in urban versus rural areas The study area includes a total of 368 ZIP codes compared to a total of 2371 block groups ZIP code average population in the study area is approximately 8000 compared to block group average population of 1200 Comparison of Figures 2 and 3 shows that using population-weighted centroid distance to the river produces different exposure areas than adjacency Fifty-three (20) of the adjacent block groups were categorized for the distance indicator as distant because the population in the block group was concentrated in the area farther than one mile from the river On the other hand about 11 of the 2116 non-adjacent block groups were classified as closest (77) or close (162)

Hospitalization Data

Hospitalizations for ischemic heart disease (IHD) and cerebrovascular disease (stroke) were included in this study More specifically hospital discharge records that listed as the principal diagnosis either IHD (ICD-9-CM 410-414 and 4297) or cerebrovascular disease (ICD-9-CM 430-436) and subsets of each of these categories similar to the subsets evaluated in the Sergeev et al and Shcherbatykh studies were included The subsets are AMI and angina subsets of IHD and ischemic and hemorrhagic stroke which are subsets of cerebrovascular disease (stroke) (Principal diagnosis represents the illness for which the person was admitted to the hospital) Angina is an outcome not evaluated in the prior studies It was added here for comparative purposes because it represents an outcome appropriately addressed by primary outpatient rather than inpatient hospital care Angina is one of a group of outcomes included in the recently developed national ldquoPrevention Quality Indicatorsrdquo (USDHHS AHRQ 2007) as an indicator for geographic areas or populations for whom primary care is inadequate Table 3 lists the specific ICD-9-CM codes and definitions included in this investigation

While one of the previously published studies (Huang et al 2006) showed elevations in hypertension (high blood pressure) rates this outcome is rarely listed as the principal reason for hospitalization As a secondary diagnosis however hypertension is extremely common with substantial overlap with the primary diagnoses of IHD and stroke In addition hypertension hospitalizations were not as elevated in the ZIP codes adjacent to the Hudson River as the IHD and stroke outcomes As a practical matter geocoding the hypertension hospitalization data (1364000 hospitalizations) would overwhelm the resources for this project For these reasons hypertension was not included in this follow-up investigation

The source of the hospitalization data was the NYS DOH Statewide Planning and Research Cooperative System (SPARCS) established in 1979 to collect detailed records on discharges from hospitals located in New York State Persons seen in the emergency department but not admitted are not included in the SPARCS data Hospitalizations at US Veterans Administration

10

Hospitals are also not included in this dataset We identified the hospital transfers in the data and excluded them so that individuals are only counted once for the same event

The NYS DOH Data Protection Review Board for SPARCS hospitalization data approved our use of hospitalization records for this investigation and provided approximately 460000 records with a principal diagnosis of either IHD or stroke Because of the large number of cardiovascular hospitalizations and this studyrsquos requirement for geocoding each hospitalized personrsquos address we selected a 50 random sample of the hospital discharges with principal diagnoses of IHD or stroke for inclusion in the study (229948 records) Approximately one percent of these records (2737) were eliminated from the study because they contained no residential address resulting in 227211 hospitalization records 24939 of these records were eliminated after geocoding because they were not within the 12 county study area This left 202272 records remaining in the study

For geocoding to the block group level the hospitalization street addresses were modified as needed using US Postal Service standards The addresses were assigned geographic coordinates using commercially available geocoding software (MapMarker 2004) A combination of land parcel data and street centerline files were used to assign geographic coordinates No contact was made with cases parents legal guardians or next of kin of cases to determine residential locations

Once geographic coordinates were assigned to cases through address-matching the case locations were overlaid onto digital maps of the study area using a geographic information system so that the number of observed cases falling within the study area boundaries could be determined and a 2000 Census block group could be assigned to each hospitalization In order to protect confidentiality no maps of individual case locations are provided Geographic coordinates were not able to be assigned for some addresses for example rural routes or post office boxes Internet searches were conducted for addresses that were located in facilities such as nursing homes and trailer parks

While fewer than one percent of records were missing sex or age information approximately 15000 records seven percent were missing race information Evaluation of the geographic and temporal distribution of these records showed they were not randomly geographically distributed 47 of the records with no race information were from one hospital for that hospital race was missing from 100 of the records in all years except one after 1997 mid-way through the study time-frame Evaluation of other records missing race showed similar patterns of 100 of records missing race in specific years In order to prevent biased hospitalization rate estimates records missing race were assigned race randomly assuming a race distribution equal to that of the individualrsquos residential block group

Statistical Analysis

This investigation used negative binomial multivariable regression models the statistical methods used in the previous studies (Sergeev and Carpenter 2005 [IHD] Shcherbatykh et al 2005 [stroke]) The follow-up studyrsquos analyses compared hospitalization rates for populations in block groups adjacent or in close proximity to the Hudson River to rates for populations more

11

distant from the Hudson River while adjusting for the potential influence of other factors Sex age and race-specific hospitalization count data comprise the dependent variable and the statistical methods attempt to predict the counts using the sex age and race distributions from Census data as well as block group-level demographics such as median income Models were first estimated using basic Census information for age sex and race Then more fully adjusted models were estimated using information at the block group level for income education population density Hispanic ethnicity and distance to the nearest hospital

More specifically each hospitalized personrsquos block group was used to assign quartile categories of block group median household income of population with less than a college education population density of the population identified as Hispanic and distance from the block group centroid to the nearest hospital These variables were selected in advance The population density variable was selected to account for potential effects on hospitalization rates of large differences in population density from the relatively rural northern counties to the more urban southern counties in the study area of the population identified as Hispanic was included because Hispanic ethnicity may be associated with CVD risk but Hispanic ethnicity was not available in the hospitalization records Distance to the hospital was included because studies show that shorter distances increase the likelihood that individuals with AMI and stroke survive to be admitted to the hospital

As described previously the literature about risk factors for CVD shows strong associations between lower income and education and higher CVD risk Regarding the effects of income and education on hospitalization rates the literature from analytic studies specifically of hospitalizations for CVD is quite sparse Most analytic studies focus on prevalence incidence or mortality in population-based studies or cohort studies and there are not many of these An example of a study that was useful as background is a study of income level and CVD mortality that showed strong associations of lower income and higher CVD mortality for men and women with the relationship for men being linear throughout the range of income levels while for women the relationship was only strong at levels of income below the median (Rehkopf 2008)

This follow-up studyrsquos approach was to use the same set of variables for the more fully adjusted models using quartile cut points for all the sexrace groups and types of CVD diagnostic subsets The 2000 Census data were used to create the population-weighted quartile categories to be used in the statistical analyses The quartile cut points are shown below

population density(persons per square mile)

median 1999 household income

with less than a college education

with Hispanic ethnicity

Hospital distance

Quartile 1 61 - lt507 $2499-lt$40968 0 - lt31 0 - lt2 lt25 (median) Quartile 2 507 - lt102 $40968 - lt54650 31 - lt43 2 - lt5 25 - lt6 (50-75) Quartile 3 102 - lt6410 $54650 - lt72944 43 - lt55 5 - lt10 6 ndash 11 (75-90) Quartile 4 6410 - 94372 $72944 - 200001 55 - 100 10 - 86 11 ndash 45 (90-100)

SAS version 9 provided the statistical programs used to estimate hospitalization rates for exposure proxy categories and demographic categories The models provide ldquocontrast

12

estimatesrdquo ie rate ratios for one category versus a reference category of a variable and 95 confidence intervals (CI) for these rate ratios For the location categories rate ratios for the areas adjacent are compared to areas not adjacent to the river (reference) and areas closest and close to the river are compared to areas distant from the river (reference) If a rate ratio is greater than 100 the model has estimated an elevation of hospitalizations in that category of the study population compared to the reference population If the rate ratio is less than 100 then there are fewer than expected hospitalizations compared to the reference population

The magnitude of the elevation or deficit is also estimated by the rate ratio For instance if the hospitalization rate is twice as high in the adjacent population it would result in a rate ratio of 200 while a 50 higher rate would result in a rate ratio of 150 If the hospitalization rate in the comparison group was only half as big as in the reference group this would result in a rate ratio of 050 The 95 confidence interval assists with determining whether the observed differences are statistically meaningful ie statistically significant The confidence interval provides the range in which there is a 95 probability of including the true rate ratio

In addition to directly assess potential confounding we examined the associations between income and exposure proxy categories and between income and CVD hospitalization risk Because the modeling process provides rate ratios that provide comparisons among groups but not the rates themselves we also calculated the rates per population for each sex race and income category in order to directly observe the relative magnitudes of CVD hospitalization rates Finally analyses stratified by income quartile were also conducted in order to assess the potential role of confounding by income The stratified analyses consist of negative binomial regressions conducted separately for each quartile of income

4 RESULTS

Demographics Associations with Proximity to the River

Study Population Table 4 provides detailed demographic characteristics for the total study population and the populations of the adjacent versus non-adjacent and nearer versus farther from the river block groups from the 2000 Census The total population of the study area for this one year of the study time period is 2924631 persons The first row of Table 4 shows that approximately 11 percent of the study area population is classified as living in block groups adjacent to the Hudson River and 89 live in non-adjacent block groups Approximately 8 of the studyrsquos population lives in the closest block groups approximately 10 lives in close block groups and approximately 82 in distant block groups The table also shows that approximately 81 of the total study population is identified as white 9 black 3 Asian less than 01 Pacific Islander 02 American Indian 4 ldquootherrdquo and 2 ldquomultiple racesrdquo

The 2000 Census data (Table 4) also show that the populations in block groups adjacent and nonshyadjacent to the river are similar in terms of the percent of minority population but the adjacent block groups include a smaller percentage of persons who have attended college and a higher percentage of households below the poverty level Median income is $47320 for block groups adjacent versus $62668 for block groups not adjacent to the river Data for the study area from the 1990 Census show similar distributions (data not shown)

13

For the distance to the river categories there is a gradient for the three levels of distance from the river with the closest block groups on average having higher percentages of minorities smaller percentages of persons who have attended college and higher percentages of households below the poverty level (Table 4) The populations of the close and closest block groups are more similar to each other than to the populations of the distant block groups The population of the closest block groups is estimated to be 19 black close block groups 14 black and distant block groups 8 black The proportions for the ldquootherrdquo category used in the analyses which includes multiple races and American Indians are approximately 12 in both the closest block groups and in the close block groups compared to 5 in the block groups most distant from the river Median income for the closest block groups is $40748 close block groups $44962 and most distant block groups $64847

The bar chart below illustrates the correlations between US Census 2000 indicators of block group socio-economic status and block group centroid (population weighted) distance from the river for the study population

2000 Census Demographics for Study Population by Distance to River Category

black other hispanic no college below poverty

of pop ulati on

0

10

20

30

40

50

60

lt12 mile

12 to 1 mile

gt1 mile

Table 5 shows selected demographic information for the person-year estimates used in the study analyses for the 1990-2005 study period Person-year data represent individuals from Census estimates counted repeatedly for each year of the study timeframe As described in the Methods section the estimation of person years used interpolation to account for population changes from 1990-2000 and techniques for taking account of block boundary changes The person-years estimates for demographic categories (complete data not shown) show similar patterns as the 1990 Census data (data not shown) and 2000 Census data (Table 5) Table 5 for example shows nearly identical percentages of population in each of the river distance categories as Table 4

Table 5 also shows person year information for income quartiles and exposure proxy categories as well as race and exposure proxy categories that is consistent with the Table 4 data for the

14

2000 population Table 5 row percentages show for example that while approximately 11 of the total study population live in adjacent block groups approximately 19 of the population in the lowest income quartile live in adjacent block groups By design approximately 25 of the total study population is in each of the income quartiles Table 5 column percentages show however that in the adjacent block groups 41 of the population is in the lowest income quartile and only 9 are in the highest income quartile

For the block groups within frac12 mile of the river the distribution among income quartiles is even more skewed towards the lowest income quartile 61 of the population living within frac12 mile of the river is in the lowest income quartile and only 6 is in the highest income quartile Given that approximately 25 of the total study population is within each income quartile these percentages indicate that living within the block groups closest to the river is associated with more than a doubling of the likelihood of being in a block group in the lowest socio-economic status quartile 436 of those living in the block groups close to the river (12 to one mile from the river) are in block groups in the lowest income quartile While less than a doubling this indicates that living in the close block groups is associated with a 75 increase in the likelihood of being in the lowest quartile for socio-economic status

Hospitalizations Table 6 shows the process of removing hospitalization records from the study for a variety of reasons The first column includes the 202272 records that contained a residential address and were located within the 12-county study area as described in the Methods section The next five columns show the records remaining after data cleaning exclusion of transfers elimination of records with missing covariates and exclusions due to poor geocoding results Approximately 33000 records were eliminated because they were transfers from one hospital to another for the same event For approximately 1100 records geocoding was inadequate for assigning block group for imputing race as described in the Methods section 336 records were missing sex (11) or age (325) Of the 167733 records remaining 154220 (92) were able to be adequately geocoded to a street address for assignment in a specific block group While approximately eight percent of hospitalizations were eliminated due to inadequate geocoding Table 6 shows that the percentages of hospitalization records for white black Asian and ldquootherrdquo categories and the distribution among diagnostic categories change very little after these records were dropped

The final column of Table 6 shows that of the 154220 hospitalizations in the analysis 101012 IHD hospitalizations comprise the largest health outcome category and are 65 of the hospitalizations in the study AMI hospitalizations (35811) account for 35 of the IHD hospitalizations and angina (4731) accounts for five percent of the IHD hospitalizations Stroke hospitalizations (53208) comprise 34 of the studyrsquos hospitalizations Approximately 36 of the stroke hospitalizations are for ischemic stroke (19371) and approximately 11 are for hemorrhagic stroke (5995)

Table 7 shows the distributions of hospitalizations among the four income quartiles While each income quartile (by definition) contains approximately 25 of the study population for all types of hospitalizations we see that the lowest income quartile includes more than 25 of hospitalizations from a minimum of 286 (IHD) to a maximum of 356 of hospitalizations (angina) The highest income quartile includes fewer than 25 of hospitalizations for each CVD

15

category ranging from 182 for angina to 224 for hemorrhagic stroke These deviations from the 25 quartile distribution show a strong linear pattern with the largest percentage of hospitalizations coming from residents of block groups in the lowest income quartile and the smallest percentages coming from residents of the most wealthy block groups For AMI the lowest income quartilersquos share of hospitalizations is 50 higher than the highest income quartilersquos share of hospitalizations (30 vs 20) IHD and stroke show similar differences and angina shows a doubling with the lowest income quartile contributing 36 of hospitalizations compared to the highest income quartilersquos 18 share of hospitalizations

Table 7 also shows the distribution of hospitalizations among the race categories and the study population race category distribution for comparison Blacks show lower percentages of IHD and AMI than their population percentages and higher percentages of angina and stroke While full presentation of race and sex findings is beyond the scope of this report the lower than expected percentages of IHD and AMI among blacks who on average have lower socioshyeconomic status and are at greater risk for CVD hospitalizations raised questions that were addressed by calculating hospitalization rates per 1000 population for each sex and race group for each income quartile (Appendix B Table 2) Figures 4 5 and 6 show these rates for the white black and Asian categories (These rates for the ldquootherrdquo category are comparatively very high and are not shown on these graphs This issue is addressed in the Discussion section) Figures 4 5 and 6 show the consistent and strong effects of income quartile on hospitalization rates for AMI and stroke for whites blacks and Asians of both sexes The other outcomes show similar patterns (The ldquootherrdquo group showed an opposite pattern [See Discussion Section])

Demographics Summary

The strong associations of lower income with residence closer to the river (Table 5) and lower income with higher hospitalization rates (Tables 7 amp 8) create a difficult challenge for this investigation The multivariable analyses of hospitalization rates attempt to control for socioshyeconomic differences However because these income differences which strongly affect hospitalization rates are also strongly correlated with the gradations of exposure (river distance) it will be difficult to disentangle the influence of socio-economic status from the influence of potential exposures

Statistical Analyses Proximity to the River and CVD Hospitalization Risk

All the variables described above were included in the multivariable regression models and almost every quartile category for every variable showed statistically significant and expected types of effects for all the CVD hospitalization categories (Appendix B Tables 1a and 1b) The following presentation of results focuses on the rate ratios for the exposure proxy variables Table 8a (IHD) and 8b (stroke) show the regression modeling results Results are shown first for models that include only age sex and race in addition to the exposure proxy categories and second for models that also include median income education population density Hispanic ethnicity and distance to nearest hospital The tables provide numbers of hospitalizations and person-years to assist with interpretation For example for total IHD we can see that while 108 of the study population lives in adjacent block groups 122 of the hospitalizations are among people in adjacent block groups This difference of 14 percentage points is the basis for

16

the finding of a statistically significantly elevated rate ratio (risk) for IHD among people living in the adjacent block groups (RR=133 CI 112-157)

Looking across the columns of Tables 8a and 8b we can see the rate ratios for each of the outcomes for the adjacent versus non-adjacent block groups and the results are similar to those for IHD The elevated rate ratios range from 123 for angina to 136 for AMI The only rate ratio not elevated in these first models adjusting only for age sex and race is for hemorrhagic stroke Going down the rows of the Table the next set of rate ratios are for the distance to the river categories The findings are very similar to those for the adjacent block groups Here however we have two exposure proxy categories allowing comparison of the two These rate ratios show no evidence that living in the block groups closest to the river (lt12 mile to the river) produces greater risk than living in the block groups close to the river (12 to 1 mile from the river)

The next set of rate ratios are for models with additional adjustment for the demographics of each hospitalized personrsquos block group as well as distance to the nearest hospital As expected adjustment for block group median income and other demographics accounts for some of the variability in hospitalization counts and thereby reduces the magnitude of the rate ratios estimated for the river adjacency and distance categories The rate ratio elevation for total IHD for example is reduced from 133 to 107 (RR=107 CI 104-111) The elevated rate ratios in these more fully adjusted models for the adjacency analyses are of similarly small magnitude ranging from 106 for AMI to 110 for ischemic stroke Four of the six outcomes show statistically significant elevations Again only hemorrhagic stroke shows no evidence of an elevated rate ratio

For the river distance categories none of the rate ratios show any evidence that living in the block groups closest to the river (lt12 mile to the river) produces greater risk than living in the block groups close to the river (12 to 1 mile) Of the six types of outcomes only two show statistically significantly elevated rate ratios for living in the closest block groups (total stroke and ischemic stroke) For the close block groups four out of six outcomes show statistically significantly elevated rate ratios The statistically significantly elevated rate ratios vary from 106 (AMI stroke) to 113 (angina)

To address the strong evidence described earlier for confounding in these data ie the association of lower income with both CVD hospitalizations and residence closer to the river we also conducted the analyses separately for each quartile of income These results are presented in Table 9 The first results are from analyses adjusting only for age sex and race and the second set of results are from models including the additional demographic factors For simplicity rate ratios are shown only for block groups closest to the river compared as in the other models to block groups farther than one mile from the river (Results were similar for the other distance categories (data not shown)

For the two lowest income quartiles Table 9 shows results similar to those for the study population as a whole (Table 8a 8b) However for the two highest income quartiles which represent frac12 of the study population Table 9 shows a consistent pattern of reduced rate ratios for residence in the closest block groups within frac12 mile of the river In the highest income quartile

17

(PCBs) polychlorinated dibenzo-p-dioxins (PCDDs) polychorinated dibenso-furans (PCDFs) and pprsquo-diphenyldichloroethene (DDE) a breakdown product of the pesticide dichlorodiphenyltrichloroethane (DDT) These chemicals are stored in human fat tissues accumulating over an individualrsquos lifetime They also accumulate in fish and other species thereby concentrating up the food chain Levels of most POPs measured in blood serum of Americans have decreased in the last few decades An exception is polybrominated diphenyl ethers (PBDEs) levels of which have been increasing in recent years While levels of many POPs are declining a large number continue to be detectable in blood for most of the US population (USDHHS 2009)

This follow-up investigation focuses on the Hudson River where the POPs of concern are PCBs PCBs are a group of synthetic chlorinated compounds manufactured in the US through 1977 Production was halted due to concerns about the persistence of PCBs in the environment and the toxicity of PCBs shown in animal models PCBs were used primarily by the electrical industry for capacitors and transformers and were also used in hydraulic fluids fluorescent light fixtures flame retardants inks adhesives carbonless copy paper paints pesticide extenders plasticizers and wire insulators PCBs can be oily liquids or solids and some are volatile and can exist as vapor in air PCBs in the environment are generally mixtures of the many different types known as congeners (ATSDR 2000)

The variety of types of PCBs adds to the complexity of studying adverse animal and human health effects Increasingly PCB research focuses on specific PCB congeners and groups of congeners because specific types of congeners appear to have different mechanisms of action and toxicity Dioxin-like PCBs (DLPCBs) are an important subset of PCBs due to concerns about their toxicity The focus on dioxin-like PCBs is paralleled by increasing research on the group of POPs with dioxin-like properties This type of PCB and POP research assigns toxic potency values relative to dioxin to various PCBs and POPs to provide an overall toxic equivalency value to specific PCB and other POPs

From 2002 through 2007 NYS DOH and SUNY Albany School of Public Health researchers conducted the Hudson River Communities Study among residents of Hudson Falls and Fort Edward the villages where the PCB discharges from two GE electrical capacitor plants occurred The comparison community was upriver from the discharges in Glens Falls This studyrsquos evaluation of exposures included measuring PCBs in outdoor air and in blood serum The study found that PCB levels in outdoor air were statistically significantly higher in the Hudson FallsFort Edward area (072 nanograms per cubic meter) than in the upriver Glens Falls area (040 nanograms per cubic meter) but that the PCB concentrations in the Hudson FallsFort Edward area were lower than those in other communities with known PCB-contaminated sites and similar to levels reported in other locations in the northeastern United States A statistically significant reduction in the total PCB concentrations in outdoor air samples was observed at a distance of 1200 meters (34 mile) from the river from 0760 ngm3 to 0497 ngm3 (Fitzgerald et al 2007 Palmer et al 2008)

The average PCB concentrations in serum for the study and comparison populations did not differ significantly (307 ppb wet weight versus 323 ppb) The average toxic equivalency quantities (TEQ) for dioxin-like PCB congeners for the study and comparison area populations

6

also did not differ In addition the study found no detectable differences in study subjectsrsquo serum PCB levels associated with proximity to contaminated portions of the river or other PCB sources The study did find that lifetime consumption of Hudson River fish was associated with higher serum PCB levels (Fitzgerald 2007 Palmer 2008) Another recent biomonitoring study of Brooklyn Manhattan and New Jersey fisherman who consumed versus those who did not consume their catch from the lower Hudson River and New YorkNew Jersey harbor suggests eating fish from these waters is not associated with an increased body burden of PCBs or most other organochlorines This study did not address the inhalation pathway (Morland 2008)

PCBs and Cardiovascular Disease Outcomes

Research studies on adverse health effects associated with PCB exposure have focused on many types of outcomes organs and biological systems including cancer liver reproductive and developmental effects thyroid endocrine immune system and neurological effects as well as specific effects on the skin and eyes Relatively little research has focused to date on associations between PCB exposures and CVD outcomes The majority of human studies have assessed health effects in occupational groups having relatively high-level exposures among people consuming fish from contaminated waters or people exposed to relatively high levels due to accidental contamination of food (USDHHS NIH 2010)

PCBs have been shown to produce a variety of toxic effects in animals These effects include cancer allergies hypersensitivity damage to the central and peripheral nervous systems reproductive disorders and disruption of the immune system (ATSDR 2000) In contrast research to date does not tell us whether specific types of human health effects have occurred from the low-level exposures experienced by the general population Some human epidemiological studies of relatively low levels of PCDDs PCDFs and PCBs suggest possible relationships not only with neurodevelopment of infants but also diabetes and a condition called insulin resistance conditions that increase CVD risk (Longnecker 2001 Carpenter 2006 Arisawa 2005)

In 1998 a study of a population in Seveso Italy exposed to relatively pure TCDD (dioxin) from an industrial accident showed increased mortality from CVD respiratory diseases and diabetes (Bertazzi 1998) These findings suggested the possibility that other dioxin-like chemicals such as dioxin-like PCBs might increase risk for these diseases Some more recent animal studies described below suggest possible mechanisms of action for PCB effects that increase risk for CVD diabetes and hypertension Another relevant source of information is from research using the biological monitoring data from the National Health and Nutrition Examination Survey (NHANES) project to look at the association between PCB levels in blood and self-reported diabetes and heart disease The body of evidence based on analytical human epidemiological studies remains relatively sparse however

Some recent studies using NHANES biomonitoring data show associations between PCB levels and CVD (Lee 2006 Lee 2007 Ha 2007 Everett 2008) These studies are cross-sectional (comparisons at one point in time) evaluations of PCB levels measured in blood samples from individuals and self-reported CVD and diabetes a risk factor for CVD While these studies use individual-level biomonitoring and self-reported health outcome data they are considered to be

7

relatively weak in terms of providing evidence of causality because both exposure and outcome are measured at the same point in time Given the nature of NHANES data collection and reporting practices these studies are not able to control for possible geographical confounding which may occur if urban versus rural residence is associated with higher PCB levels and higher risk for CVD for example The higher PCB levels among African Americans in these data suggest urban residence may be playing a role in the PCB data

The results of animal studies suggest multiple mechanisms by which PCB exposures could affect cardiovascular risk as well as multiple types of adverse health outcomes One hypothesized mechanisms of action specifically for dioxin-like PCBs is that they may have anti-estrogenic effects which impact CVD risk factors such as serum cholesterol levels and blood pressure (Lind 2004) Several studies suggest PCBs interact with a specific receptor (AhR) and increase cellular oxidative stress causing inflammation cell injury and cardiovascular dysfunction (Hennig 2002) Similarly some studies suggest PCB-induced inflammation of fat cells leads to obesity and increased risk for coronary artery disease (atherosclerosis) (Arsenescu 2008)

3 FOLLOW-UP STUDY DATA SOURCES AND METHODS

This follow-up investigation will provide perspective on the 2004-2005 findings (Sergeev and Carpenter 2005 Shcherbatykh et al 2005) by producing results from analyses conducted at a more detailed level of geography the block group rather than ZIP code in order to better control for confounding When the term confounding is used by epidemioogists it refers to the problem that can arise when there is a factor such as income level that is not the factor of interest in the study but that is related to the risk factor being studied (residential location ndash proxy for exposure) and to the disease being studied (CVD) This type of factor is labeled a ldquoconfounderrdquo because it can confound ie interfere with or confuse the evaluation of associations among the variables of interest in the data All types of epidemiological studies face issues of confounding but ecological ie group-level analyses face additional challenges related to potential confounding because analyses can not account for some important individual differences within the groups that are the units of analysis (Morganstern 1998b Hertz-Picciotto 1998) This investigation was able to control for each individualrsquos age sex and race but not for individual-level risk factors such as income or education level

The follow-up investigationrsquos specific objectives are to

improve the ability to control for confounding by socio-economic status through the use of block groups rather than ZIP codes as the unit of analysis

compare the hospitalization rates among residents of block groups nearest the Hudson River to those among residents of block groups farther from the Hudson River

compare the findings of this follow-up study to those of the earlier studies

Study Area and Exposure Areas

The overall study area is comprised of the 12 counties with geographic areas abutting the Hudson River south of the source of PCB contamination from the General Electric facility in Hudson Falls The 12 counties are Westchester Rockland Putnam Orange Dutchess Ulster

8

Columbia Greene Rensselaer Albany Washington and Saratoga US Census boundaries are used to define the study area because census population estimates provide the data for the study area populationrsquos size and the populationrsquos age sex and race distribution Because the study period spanned 1990 through 2005 and populations change over time 1990 and 2000 Census data were used to linearly interpolate the between-year populations from 1991 through 1999 for each block group 2000 data were used without extrapolation to estimate block group populations from 2001 through 2005

Because block group boundaries change between the Census years 1990 block populations were normalized to fit 2000 block group boundaries before the estimation of sex age and race group specific populations In the digital map 2000 block group boundaries were overlaid on the 1990 block centroids to assign 1990 block population counts to the corresponding 2000 block group boundaries Then 1990 block populations were summed to fit the block group boundaries Next population counts by age group (0-14 15-24 25-34 35-44 45-54 55-64 65-74 75+) sex (male female) and race (white black asian other) were estimated for each year at the block group level This investigation includes all ages over 25 years and the study population is divided into four race categories white black or African-American Asian and ldquoOtherrdquo The Pacific Island group was combined with the Asian group and American Indian ldquootherrdquo and ldquomultiple racesrdquo categories are combined into a larger ldquootherrdquo category

Classification of the population into potential exposure categories for this study is based exclusively on location relative to the Hudson River Two types of exposure proxy categories were developed block group adjacency and block group centroid distance to the river shore For the adjacency categories block groups were assigned to the adjacent category if they directly abutted the Hudson River and all other block groups in the 12 counties were assigned to the nonshyadjacent category

For the distance categories a central point (centroid) within each block group was identified and used to assign block group centroid distance to the river shore The block group centroids were population-weighted which means the geographic distribution of households within 2000 census blocks were used to locate block group centroids based on the geographic population distribution By locating a population-weighted block centroid rather than a more straightforward distance centroid the assignment of block group distance to the river shore more accurately reflects distance to the river for the majority of block group residents Buffers of one-half mile and one mile from the Hudson River boundary were created for the assignment of each block group into one of the three distance categories less than frac12 mile (closest) frac12 to one mile (close) and farther than one mile from the river (distant)

Distance categories were selected a priori at the start of the study Because environmental sampling data were not collected along the length of the river included in this investigation we are not able to determine whether these distance categories represent actual differences in exposures As discussed previously a statistically significant reduction in the total PCB concentrations in outdoor air samples was observed at a distance of 1200 meters (34 mile) from the river from 0760 ngm3 to 0497 ngm3 in a study conducted in the Hudson Falls Fort Edward area near the source of PCBs (Fitzgerald 2007 Palmer 2007)

9

For the adjacency categories of the 2371 block groups in the study area 255 (11) were classified as adjacent to the Hudson River and 2116 (89) were classified as non-adjacent (Figure 2) For the distance categories 225 (9) were classified as closest to the river 216 (9) as close and 1930 block groups (81) as distant from the river

Comparison of Figures 1 and 2 show that block groups include smaller geographic areas than ZIP codes and that block groups are relatively smaller in area in urban versus rural areas The study area includes a total of 368 ZIP codes compared to a total of 2371 block groups ZIP code average population in the study area is approximately 8000 compared to block group average population of 1200 Comparison of Figures 2 and 3 shows that using population-weighted centroid distance to the river produces different exposure areas than adjacency Fifty-three (20) of the adjacent block groups were categorized for the distance indicator as distant because the population in the block group was concentrated in the area farther than one mile from the river On the other hand about 11 of the 2116 non-adjacent block groups were classified as closest (77) or close (162)

Hospitalization Data

Hospitalizations for ischemic heart disease (IHD) and cerebrovascular disease (stroke) were included in this study More specifically hospital discharge records that listed as the principal diagnosis either IHD (ICD-9-CM 410-414 and 4297) or cerebrovascular disease (ICD-9-CM 430-436) and subsets of each of these categories similar to the subsets evaluated in the Sergeev et al and Shcherbatykh studies were included The subsets are AMI and angina subsets of IHD and ischemic and hemorrhagic stroke which are subsets of cerebrovascular disease (stroke) (Principal diagnosis represents the illness for which the person was admitted to the hospital) Angina is an outcome not evaluated in the prior studies It was added here for comparative purposes because it represents an outcome appropriately addressed by primary outpatient rather than inpatient hospital care Angina is one of a group of outcomes included in the recently developed national ldquoPrevention Quality Indicatorsrdquo (USDHHS AHRQ 2007) as an indicator for geographic areas or populations for whom primary care is inadequate Table 3 lists the specific ICD-9-CM codes and definitions included in this investigation

While one of the previously published studies (Huang et al 2006) showed elevations in hypertension (high blood pressure) rates this outcome is rarely listed as the principal reason for hospitalization As a secondary diagnosis however hypertension is extremely common with substantial overlap with the primary diagnoses of IHD and stroke In addition hypertension hospitalizations were not as elevated in the ZIP codes adjacent to the Hudson River as the IHD and stroke outcomes As a practical matter geocoding the hypertension hospitalization data (1364000 hospitalizations) would overwhelm the resources for this project For these reasons hypertension was not included in this follow-up investigation

The source of the hospitalization data was the NYS DOH Statewide Planning and Research Cooperative System (SPARCS) established in 1979 to collect detailed records on discharges from hospitals located in New York State Persons seen in the emergency department but not admitted are not included in the SPARCS data Hospitalizations at US Veterans Administration

10

Hospitals are also not included in this dataset We identified the hospital transfers in the data and excluded them so that individuals are only counted once for the same event

The NYS DOH Data Protection Review Board for SPARCS hospitalization data approved our use of hospitalization records for this investigation and provided approximately 460000 records with a principal diagnosis of either IHD or stroke Because of the large number of cardiovascular hospitalizations and this studyrsquos requirement for geocoding each hospitalized personrsquos address we selected a 50 random sample of the hospital discharges with principal diagnoses of IHD or stroke for inclusion in the study (229948 records) Approximately one percent of these records (2737) were eliminated from the study because they contained no residential address resulting in 227211 hospitalization records 24939 of these records were eliminated after geocoding because they were not within the 12 county study area This left 202272 records remaining in the study

For geocoding to the block group level the hospitalization street addresses were modified as needed using US Postal Service standards The addresses were assigned geographic coordinates using commercially available geocoding software (MapMarker 2004) A combination of land parcel data and street centerline files were used to assign geographic coordinates No contact was made with cases parents legal guardians or next of kin of cases to determine residential locations

Once geographic coordinates were assigned to cases through address-matching the case locations were overlaid onto digital maps of the study area using a geographic information system so that the number of observed cases falling within the study area boundaries could be determined and a 2000 Census block group could be assigned to each hospitalization In order to protect confidentiality no maps of individual case locations are provided Geographic coordinates were not able to be assigned for some addresses for example rural routes or post office boxes Internet searches were conducted for addresses that were located in facilities such as nursing homes and trailer parks

While fewer than one percent of records were missing sex or age information approximately 15000 records seven percent were missing race information Evaluation of the geographic and temporal distribution of these records showed they were not randomly geographically distributed 47 of the records with no race information were from one hospital for that hospital race was missing from 100 of the records in all years except one after 1997 mid-way through the study time-frame Evaluation of other records missing race showed similar patterns of 100 of records missing race in specific years In order to prevent biased hospitalization rate estimates records missing race were assigned race randomly assuming a race distribution equal to that of the individualrsquos residential block group

Statistical Analysis

This investigation used negative binomial multivariable regression models the statistical methods used in the previous studies (Sergeev and Carpenter 2005 [IHD] Shcherbatykh et al 2005 [stroke]) The follow-up studyrsquos analyses compared hospitalization rates for populations in block groups adjacent or in close proximity to the Hudson River to rates for populations more

11

distant from the Hudson River while adjusting for the potential influence of other factors Sex age and race-specific hospitalization count data comprise the dependent variable and the statistical methods attempt to predict the counts using the sex age and race distributions from Census data as well as block group-level demographics such as median income Models were first estimated using basic Census information for age sex and race Then more fully adjusted models were estimated using information at the block group level for income education population density Hispanic ethnicity and distance to the nearest hospital

More specifically each hospitalized personrsquos block group was used to assign quartile categories of block group median household income of population with less than a college education population density of the population identified as Hispanic and distance from the block group centroid to the nearest hospital These variables were selected in advance The population density variable was selected to account for potential effects on hospitalization rates of large differences in population density from the relatively rural northern counties to the more urban southern counties in the study area of the population identified as Hispanic was included because Hispanic ethnicity may be associated with CVD risk but Hispanic ethnicity was not available in the hospitalization records Distance to the hospital was included because studies show that shorter distances increase the likelihood that individuals with AMI and stroke survive to be admitted to the hospital

As described previously the literature about risk factors for CVD shows strong associations between lower income and education and higher CVD risk Regarding the effects of income and education on hospitalization rates the literature from analytic studies specifically of hospitalizations for CVD is quite sparse Most analytic studies focus on prevalence incidence or mortality in population-based studies or cohort studies and there are not many of these An example of a study that was useful as background is a study of income level and CVD mortality that showed strong associations of lower income and higher CVD mortality for men and women with the relationship for men being linear throughout the range of income levels while for women the relationship was only strong at levels of income below the median (Rehkopf 2008)

This follow-up studyrsquos approach was to use the same set of variables for the more fully adjusted models using quartile cut points for all the sexrace groups and types of CVD diagnostic subsets The 2000 Census data were used to create the population-weighted quartile categories to be used in the statistical analyses The quartile cut points are shown below

population density(persons per square mile)

median 1999 household income

with less than a college education

with Hispanic ethnicity

Hospital distance

Quartile 1 61 - lt507 $2499-lt$40968 0 - lt31 0 - lt2 lt25 (median) Quartile 2 507 - lt102 $40968 - lt54650 31 - lt43 2 - lt5 25 - lt6 (50-75) Quartile 3 102 - lt6410 $54650 - lt72944 43 - lt55 5 - lt10 6 ndash 11 (75-90) Quartile 4 6410 - 94372 $72944 - 200001 55 - 100 10 - 86 11 ndash 45 (90-100)

SAS version 9 provided the statistical programs used to estimate hospitalization rates for exposure proxy categories and demographic categories The models provide ldquocontrast

12

estimatesrdquo ie rate ratios for one category versus a reference category of a variable and 95 confidence intervals (CI) for these rate ratios For the location categories rate ratios for the areas adjacent are compared to areas not adjacent to the river (reference) and areas closest and close to the river are compared to areas distant from the river (reference) If a rate ratio is greater than 100 the model has estimated an elevation of hospitalizations in that category of the study population compared to the reference population If the rate ratio is less than 100 then there are fewer than expected hospitalizations compared to the reference population

The magnitude of the elevation or deficit is also estimated by the rate ratio For instance if the hospitalization rate is twice as high in the adjacent population it would result in a rate ratio of 200 while a 50 higher rate would result in a rate ratio of 150 If the hospitalization rate in the comparison group was only half as big as in the reference group this would result in a rate ratio of 050 The 95 confidence interval assists with determining whether the observed differences are statistically meaningful ie statistically significant The confidence interval provides the range in which there is a 95 probability of including the true rate ratio

In addition to directly assess potential confounding we examined the associations between income and exposure proxy categories and between income and CVD hospitalization risk Because the modeling process provides rate ratios that provide comparisons among groups but not the rates themselves we also calculated the rates per population for each sex race and income category in order to directly observe the relative magnitudes of CVD hospitalization rates Finally analyses stratified by income quartile were also conducted in order to assess the potential role of confounding by income The stratified analyses consist of negative binomial regressions conducted separately for each quartile of income

4 RESULTS

Demographics Associations with Proximity to the River

Study Population Table 4 provides detailed demographic characteristics for the total study population and the populations of the adjacent versus non-adjacent and nearer versus farther from the river block groups from the 2000 Census The total population of the study area for this one year of the study time period is 2924631 persons The first row of Table 4 shows that approximately 11 percent of the study area population is classified as living in block groups adjacent to the Hudson River and 89 live in non-adjacent block groups Approximately 8 of the studyrsquos population lives in the closest block groups approximately 10 lives in close block groups and approximately 82 in distant block groups The table also shows that approximately 81 of the total study population is identified as white 9 black 3 Asian less than 01 Pacific Islander 02 American Indian 4 ldquootherrdquo and 2 ldquomultiple racesrdquo

The 2000 Census data (Table 4) also show that the populations in block groups adjacent and nonshyadjacent to the river are similar in terms of the percent of minority population but the adjacent block groups include a smaller percentage of persons who have attended college and a higher percentage of households below the poverty level Median income is $47320 for block groups adjacent versus $62668 for block groups not adjacent to the river Data for the study area from the 1990 Census show similar distributions (data not shown)

13

For the distance to the river categories there is a gradient for the three levels of distance from the river with the closest block groups on average having higher percentages of minorities smaller percentages of persons who have attended college and higher percentages of households below the poverty level (Table 4) The populations of the close and closest block groups are more similar to each other than to the populations of the distant block groups The population of the closest block groups is estimated to be 19 black close block groups 14 black and distant block groups 8 black The proportions for the ldquootherrdquo category used in the analyses which includes multiple races and American Indians are approximately 12 in both the closest block groups and in the close block groups compared to 5 in the block groups most distant from the river Median income for the closest block groups is $40748 close block groups $44962 and most distant block groups $64847

The bar chart below illustrates the correlations between US Census 2000 indicators of block group socio-economic status and block group centroid (population weighted) distance from the river for the study population

2000 Census Demographics for Study Population by Distance to River Category

black other hispanic no college below poverty

of pop ulati on

0

10

20

30

40

50

60

lt12 mile

12 to 1 mile

gt1 mile

Table 5 shows selected demographic information for the person-year estimates used in the study analyses for the 1990-2005 study period Person-year data represent individuals from Census estimates counted repeatedly for each year of the study timeframe As described in the Methods section the estimation of person years used interpolation to account for population changes from 1990-2000 and techniques for taking account of block boundary changes The person-years estimates for demographic categories (complete data not shown) show similar patterns as the 1990 Census data (data not shown) and 2000 Census data (Table 5) Table 5 for example shows nearly identical percentages of population in each of the river distance categories as Table 4

Table 5 also shows person year information for income quartiles and exposure proxy categories as well as race and exposure proxy categories that is consistent with the Table 4 data for the

14

2000 population Table 5 row percentages show for example that while approximately 11 of the total study population live in adjacent block groups approximately 19 of the population in the lowest income quartile live in adjacent block groups By design approximately 25 of the total study population is in each of the income quartiles Table 5 column percentages show however that in the adjacent block groups 41 of the population is in the lowest income quartile and only 9 are in the highest income quartile

For the block groups within frac12 mile of the river the distribution among income quartiles is even more skewed towards the lowest income quartile 61 of the population living within frac12 mile of the river is in the lowest income quartile and only 6 is in the highest income quartile Given that approximately 25 of the total study population is within each income quartile these percentages indicate that living within the block groups closest to the river is associated with more than a doubling of the likelihood of being in a block group in the lowest socio-economic status quartile 436 of those living in the block groups close to the river (12 to one mile from the river) are in block groups in the lowest income quartile While less than a doubling this indicates that living in the close block groups is associated with a 75 increase in the likelihood of being in the lowest quartile for socio-economic status

Hospitalizations Table 6 shows the process of removing hospitalization records from the study for a variety of reasons The first column includes the 202272 records that contained a residential address and were located within the 12-county study area as described in the Methods section The next five columns show the records remaining after data cleaning exclusion of transfers elimination of records with missing covariates and exclusions due to poor geocoding results Approximately 33000 records were eliminated because they were transfers from one hospital to another for the same event For approximately 1100 records geocoding was inadequate for assigning block group for imputing race as described in the Methods section 336 records were missing sex (11) or age (325) Of the 167733 records remaining 154220 (92) were able to be adequately geocoded to a street address for assignment in a specific block group While approximately eight percent of hospitalizations were eliminated due to inadequate geocoding Table 6 shows that the percentages of hospitalization records for white black Asian and ldquootherrdquo categories and the distribution among diagnostic categories change very little after these records were dropped

The final column of Table 6 shows that of the 154220 hospitalizations in the analysis 101012 IHD hospitalizations comprise the largest health outcome category and are 65 of the hospitalizations in the study AMI hospitalizations (35811) account for 35 of the IHD hospitalizations and angina (4731) accounts for five percent of the IHD hospitalizations Stroke hospitalizations (53208) comprise 34 of the studyrsquos hospitalizations Approximately 36 of the stroke hospitalizations are for ischemic stroke (19371) and approximately 11 are for hemorrhagic stroke (5995)

Table 7 shows the distributions of hospitalizations among the four income quartiles While each income quartile (by definition) contains approximately 25 of the study population for all types of hospitalizations we see that the lowest income quartile includes more than 25 of hospitalizations from a minimum of 286 (IHD) to a maximum of 356 of hospitalizations (angina) The highest income quartile includes fewer than 25 of hospitalizations for each CVD

15

category ranging from 182 for angina to 224 for hemorrhagic stroke These deviations from the 25 quartile distribution show a strong linear pattern with the largest percentage of hospitalizations coming from residents of block groups in the lowest income quartile and the smallest percentages coming from residents of the most wealthy block groups For AMI the lowest income quartilersquos share of hospitalizations is 50 higher than the highest income quartilersquos share of hospitalizations (30 vs 20) IHD and stroke show similar differences and angina shows a doubling with the lowest income quartile contributing 36 of hospitalizations compared to the highest income quartilersquos 18 share of hospitalizations

Table 7 also shows the distribution of hospitalizations among the race categories and the study population race category distribution for comparison Blacks show lower percentages of IHD and AMI than their population percentages and higher percentages of angina and stroke While full presentation of race and sex findings is beyond the scope of this report the lower than expected percentages of IHD and AMI among blacks who on average have lower socioshyeconomic status and are at greater risk for CVD hospitalizations raised questions that were addressed by calculating hospitalization rates per 1000 population for each sex and race group for each income quartile (Appendix B Table 2) Figures 4 5 and 6 show these rates for the white black and Asian categories (These rates for the ldquootherrdquo category are comparatively very high and are not shown on these graphs This issue is addressed in the Discussion section) Figures 4 5 and 6 show the consistent and strong effects of income quartile on hospitalization rates for AMI and stroke for whites blacks and Asians of both sexes The other outcomes show similar patterns (The ldquootherrdquo group showed an opposite pattern [See Discussion Section])

Demographics Summary

The strong associations of lower income with residence closer to the river (Table 5) and lower income with higher hospitalization rates (Tables 7 amp 8) create a difficult challenge for this investigation The multivariable analyses of hospitalization rates attempt to control for socioshyeconomic differences However because these income differences which strongly affect hospitalization rates are also strongly correlated with the gradations of exposure (river distance) it will be difficult to disentangle the influence of socio-economic status from the influence of potential exposures

Statistical Analyses Proximity to the River and CVD Hospitalization Risk

All the variables described above were included in the multivariable regression models and almost every quartile category for every variable showed statistically significant and expected types of effects for all the CVD hospitalization categories (Appendix B Tables 1a and 1b) The following presentation of results focuses on the rate ratios for the exposure proxy variables Table 8a (IHD) and 8b (stroke) show the regression modeling results Results are shown first for models that include only age sex and race in addition to the exposure proxy categories and second for models that also include median income education population density Hispanic ethnicity and distance to nearest hospital The tables provide numbers of hospitalizations and person-years to assist with interpretation For example for total IHD we can see that while 108 of the study population lives in adjacent block groups 122 of the hospitalizations are among people in adjacent block groups This difference of 14 percentage points is the basis for

16

the finding of a statistically significantly elevated rate ratio (risk) for IHD among people living in the adjacent block groups (RR=133 CI 112-157)

Looking across the columns of Tables 8a and 8b we can see the rate ratios for each of the outcomes for the adjacent versus non-adjacent block groups and the results are similar to those for IHD The elevated rate ratios range from 123 for angina to 136 for AMI The only rate ratio not elevated in these first models adjusting only for age sex and race is for hemorrhagic stroke Going down the rows of the Table the next set of rate ratios are for the distance to the river categories The findings are very similar to those for the adjacent block groups Here however we have two exposure proxy categories allowing comparison of the two These rate ratios show no evidence that living in the block groups closest to the river (lt12 mile to the river) produces greater risk than living in the block groups close to the river (12 to 1 mile from the river)

The next set of rate ratios are for models with additional adjustment for the demographics of each hospitalized personrsquos block group as well as distance to the nearest hospital As expected adjustment for block group median income and other demographics accounts for some of the variability in hospitalization counts and thereby reduces the magnitude of the rate ratios estimated for the river adjacency and distance categories The rate ratio elevation for total IHD for example is reduced from 133 to 107 (RR=107 CI 104-111) The elevated rate ratios in these more fully adjusted models for the adjacency analyses are of similarly small magnitude ranging from 106 for AMI to 110 for ischemic stroke Four of the six outcomes show statistically significant elevations Again only hemorrhagic stroke shows no evidence of an elevated rate ratio

For the river distance categories none of the rate ratios show any evidence that living in the block groups closest to the river (lt12 mile to the river) produces greater risk than living in the block groups close to the river (12 to 1 mile) Of the six types of outcomes only two show statistically significantly elevated rate ratios for living in the closest block groups (total stroke and ischemic stroke) For the close block groups four out of six outcomes show statistically significantly elevated rate ratios The statistically significantly elevated rate ratios vary from 106 (AMI stroke) to 113 (angina)

To address the strong evidence described earlier for confounding in these data ie the association of lower income with both CVD hospitalizations and residence closer to the river we also conducted the analyses separately for each quartile of income These results are presented in Table 9 The first results are from analyses adjusting only for age sex and race and the second set of results are from models including the additional demographic factors For simplicity rate ratios are shown only for block groups closest to the river compared as in the other models to block groups farther than one mile from the river (Results were similar for the other distance categories (data not shown)

For the two lowest income quartiles Table 9 shows results similar to those for the study population as a whole (Table 8a 8b) However for the two highest income quartiles which represent frac12 of the study population Table 9 shows a consistent pattern of reduced rate ratios for residence in the closest block groups within frac12 mile of the river In the highest income quartile

17